Using constraint programming to set resource quota limits for allocating resources to consumers

Abstract

Resource allocation constraints include resource limits and resource guarantees. If a consumer uses more resources than guaranteed, it becomes vulnerable to interruptions by other consumers. Resources are designated and / or allocated to consumers based on resource limits and resource guarantees. A constraint programming (CP) solver determines resource limits and resource guarantees that minimize vulnerability and / or vulnerability costs based on resource usage data. The CP data model includes limit elements, guarantee elements, and vulnerability elements. The CP data model further includes guarantee-vulnerability constraints that depend on excess distributions that have been generated from the consumer resource usage data. The CP data model declaratively expresses the combinatorial properties of a problem in terms of constraints. CP is a type of declarative programming.

Description

[Technical field]

[0001] RELATED APPLICATIONS; INCORPORATION BY REFERENCE This application is related to U.S. Non-Provisional Patent Application No. 17 / 302,657, filed May 10, 2021, and is incorporated herein by reference.

[0002] Technical Field The present disclosure relates to the use of constraint programming, and in particular to the use of constraint programming to set resource allocation limits for allocating resources to consumers in a computing system. [Background technology]

[0003] background A computing system includes a set of resources (e.g., a central processing unit (CPU)) that are shared among a set of consumers (e.g., database instances). If one consumer places a large demand on the shared set of resources, that consumer can adversely affect the performance of other consumers.

[0004] One way to mitigate adverse effects between consumers sharing a finite set of resources is to enforce resource quota limits on consumers. However, consumers are associated with different process types, priorities, and / or other attributes. Furthermore, the process load of consumers changes dynamically. If resource quota limits are set without considering the static attributes and / or dynamic load of consumers, resource allocation may be inefficient. For example, some consumers may resist the limits while some resources are underutilized.

[0005] The approaches described in this section are approaches that could be pursued, but not necessarily approaches that have been previously conceived or pursued, and thus, unless otherwise indicated, it should not be assumed that any of the approaches described in this section qualify as prior art merely by virtue of their inclusion in this section.

[0006] The embodiments are illustrated by way of example and not by way of limitation in the figures of the accompanying drawings. It should be noted that references to "one" or "an" embodiment in this disclosure do not necessarily refer to the same embodiment, but rather they mean at least one. In the drawings: [Brief description of the drawings]

[0007] [Figure 1] FIG. 1 illustrates a resource manager for managing a computing system in accordance with one or more embodiments. [Figure 2A] FIG. 1 illustrates a constraint programming data model and search command generation system according to one or more embodiments. [Figure 2B] FIG. 1 illustrates a constraint programming data model and search command generation system according to one or more embodiments. [Diagram 3] FIG. 1 illustrates an example set of operations for updating resource limits and / or resource guarantees for consumers in a computing system based on resource usage over a particular historical period, according to one or more embodiments. [Figure 4] FIG. 1 illustrates an example set of operations for generating a constraint programming data model in accordance with one or more embodiments. [Figure 5A] FIG. 1 illustrates an example set of operations for generating constraint programming search directives, according to one or more embodiments. [Figure 5B] FIG. 1 illustrates an example set of operations for generating constraint programming search directives, according to one or more embodiments. [Figure 5C] FIG. 1 illustrates an example set of operations for generating constraint programming search directives, according to one or more embodiments. [Figure 6] FIG. 2 illustrates an example of a constraint programming data model and a set of operations for applying constraint programming search directives to a constraint programming solver, according to one or more embodiments. [Figure 7] FIG. 10 illustrates an example set of operations for assigning resources to consumers based on resource guarantees being updated by a constraint programming solver, in accordance with one or more embodiments. [Figure 8] FIG. 1 illustrates an example set of operations for queuing process requests that exceed resource limits updated by a constraint programming solver, in accordance with one or more embodiments. [Figure 9] FIG. 1 is a block diagram illustrating a computer system according to one or more embodiments. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] Detailed Description In the following description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding. One or more embodiments may be practiced without these specific details. Features described in one embodiment may be combined with features described in different embodiments. In some instances, well-known structures and devices are described in block diagram form in order to avoid unnecessarily obscuring the present invention. 1. Overview 2. Computing System Architecture 3. Constraint Programming Data Model and Search Instruction Generation System Architecture 4. Determining and Updating Resource Allocation Limits 5. Generation of constraint programming data model and constraint programming search instructions and application to constraint programming solver 6. Enforcing resource guarantees and limits 7. Hardware Overview 8.Others;Extensions 1. Overview One or more embodiments include setting resource allocation limits for allocating resources to consumers in a computing system. Resource allocation limits include resource limits and resource guarantees. A resource limit is an upper limit on the number of resources a particular consumer may use at a particular time. A resource guarantee is a lower limit on the number of resources a particular consumer is guaranteed to have access to at a particular time. A consumer that uses more resources than its guaranteed resources becomes vulnerable to interruptions in the event of a sudden increase in resource demand by other consumers. As used herein, the term "vulnerability" refers to a consumer's resource usage that exceeds the consumer's resource guarantee.

[0009] One or more embodiments include using constraint programming (CP) to set resource allocation limits for allocating resources to consumers in a computing system. CP is a type of declarative programming. CP obtains solutions to real-world problems based on the specification of a CP data model and, optionally, CP search directives. In contrast, imperative programming is based on the specification of a sequence of steps.

[0010] A CP data model includes a set of data model elements, a domain (range) of possible values ​​that can be assigned to each element, and one or more constraints that define the combinations of values ​​that can be assigned to the elements. Based on the CP data model, a set of values ​​for the set of elements that satisfies all the constraints is determined. The set of values ​​for the set of elements that satisfies all the constraints may be referred to herein as a "CP solution."

[0011] A CP search directive leads to the assignment of a set of values ​​to a set of data model elements that satisfy all constraints specified by the CP data model. A CP search directive generates a particular sequence in which value assignments to one or more data model elements are attempted. Additionally or alternatively, a CP search directive prioritizes the assignment of a particular value over other values ​​for one or more data model elements. Using different CP search directives for the same CP data model may result in different CP solutions. Additionally or alternatively, different CP search directives 234 for the same CP data model 232 may result in different efficiency levels and / or execution times for obtaining a CP solution.

[0012] One or more embodiments include generating a CP data model for determining resource allocation limits for allocating resources to consumers in a computing system. The CP data model includes several sets of data model elements. A set of limit elements represent resource limits for the consumer. A set of guarantee elements represent resource guarantees for the consumer. A set of vulnerability elements represent a number of data points indicating vulnerability (also referred to as a "vulnerability count"). A particular vulnerability element represents a number of data points indicating that a particular consumer is vulnerable with respect to a particular resource guarantee. Such number of data points is based on resource usage data for a particular consumer over a particular historical period.

[0013] The CP data model further includes several constraints. The limit-guarantee constraint relates a limit element and a guarantee element. It requires that the guarantee element (for a particular consumer) is equal to the limit element (for a particular consumer) divided by the oversubscription ratio, which is equal to the sum of the limit elements divided by the total number of resources. The total resource constraint relates a guarantee element and the total number of resources in a computing system. It requires that the sum of the guarantee elements is less than or equal to the total number of resources. The guarantee-vulnerability constraint relates a guarantee element and a vulnerability element. It requires that the vulnerability element (for a particular consumer) is equal to the number of vulnerabilities that correspond to the guarantee value that is assigned (or should be assigned) to the guarantee element (for a particular consumer) based on the resource usage over a historical period.

[0014] In one embodiment, the guarantee-vulnerability constraint is expressed as an element constraint. The element constraint accepts as input (a) an index element and (b) a set of values. The index element serves to index into the set of values ​​such that the element constraint returns a particular value from the set of values ​​that corresponds to a value that is assigned (or should be assigned) to the index element. For a guarantee-vulnerability constraint, the index element is the guarantee element (for a particular consumer) and the set of values ​​is an excess distribution (for a particular consumer). The excess distribution indicates the number of data points that indicate that the resource usage of a particular consumer exceeds a critical value, which ranges from zero to the total number of resources in the computing system. The number of such data points is based on the resource usage data of the particular consumer over a particular historical period.

[0015] One or more embodiments include generating a CP search directive to determine resource allocation limits for allocating resources to consumers in a computing system. In one embodiment, the CP search directive includes a minimization objective function associated with a vulnerability count. In one embodiment, the CP search directive is to minimize a maximum value of a set of vulnerability factors. In another embodiment, the CP search directive is to minimize an average value of a set of vulnerability factors. In another embodiment, the CP search directive is to minimize a sum of vulnerability factors (i.e., minimize the total number of vulnerabilities for the computing system). In yet another embodiment, the CP search directive includes a minimization objective function associated with a vulnerability count and a vulnerability cost, where a vulnerability cost per consumer for a particular consumer is a vulnerability factor for the particular consumer multiplied by a vulnerability factor for the particular consumer. Different consumers may be associated with different vulnerability factors. In one embodiment, the CP search directive is to minimize a maximum value of a vulnerability cost per consumer. In another embodiment, the CP search directive is to minimize an average vulnerability cost per consumer. In another embodiment, the CP search directive is to minimize the sum of the vulnerability costs per consumer (ie, minimize the total vulnerability cost for the computing system).

[0016] One or more embodiments include applying the CP data model and the CP search directive to a CP solver to obtain resource allocation limits. The resource manager obtains the updated resource limits and / or resource guarantees from the CP solver. Based on the resource guarantee for the particular consumer, the resource manager assigns resources to the particular consumer. The number of resources assigned is equal to the resource guarantee. The resource manager prohibits other consumers from using the assigned resources. Furthermore, based on the resource guarantee and / or resource limits for the particular consumer, the resource manager queues or assigns new process requests made by the particular consumer. If the number of pending process requests for the particular consumer is equal to or less than the resource guarantee, the new process request may be assigned preferentially to the assigned resources. If the number of pending process requests exceeds the resource limit, the new process request is added to the queue for the particular consumer. Thus, the resource manager ensures that the process requests assigned to the resource are within the resource limit. If the number of pending process requests exceeds the resource guarantee but is less than or equal to the resource limit, then a process request being removed from the queue of a particular consumer (or, if there is no queue, the new process request itself) may be assigned to the particular resource. In order to assign the particular resource to a process request of a particular consumer, the resource manager may interrupt an ongoing execution of a process request made by another consumer on the particular resource.

[0017] The choice of using excess distributions to model the resource allocation problem allows the CP solver to efficiently determine the optimized solution. The large amount of data embedded in the historical log of resource usage is compressed into the excess distribution, which corresponds neatly and logically to the guarantee limits via the element constraints. Furthermore, the use of the excess distributions and associated element constraints allows the CP solver to search for the optimized solution without requiring requirements on the shape of the objective function, performing derivatives of the objective function, and / or providing an initial estimate of the CP solution.

[0018] The CP data model and CP search directives allow the CP solver to efficiently determine updated resource allocation limits based on recent resource usage data. Continuously updating resource allocation limits allows a computing system to quickly adapt to changing resource usage patterns. Optimal resource allocation limits reduce the likelihood of consumer vulnerabilities and / or resource underutilization.

[0019] One or more embodiments described and / or claimed herein may not be included in this summary section.

[0020] 2. Computing System Architecture FIG. 1 illustrates a resource manager for managing a computing system according to one or more embodiments. As shown in FIG. 1, the system 100 includes a resource manager 110, a data repository 102, a set of consumers 112, a set of resources 116, and a constraint programming (CP) solver 118. In one or more embodiments, the system 100 may include more or less components than those illustrated in FIG. 1. The components illustrated in FIG. 1 may be local to each other or remote to each other. The components illustrated in FIG. 1 may be implemented in software and / or hardware. Each component may be distributed across multiple applications and / or machines. Multiple components may be combined into one application and / or machine. Operations described with respect to one component may instead be performed by another component.

[0021] In one or more embodiments, a consumer 112 refers to hardware and / or software that generates the workload of a resource 116. Examples of a consumer 112 include a database instance, a virtual machine, and / or a hardware server. The workload generated by a consumer 112 may be measured in various units, such as the number of process requests, the amount of data communicated and / or stored, and / or the duration of processing time.

[0022] In one or more embodiments, resources 116 refer to hardware and / or software that services and / or processes the workload being generated by consumers 112. Examples of resources 116 include a central processing unit (CPU), disk memory, and / or communications bandwidth.

[0023] In one or more embodiments, resource manager 110 refers to hardware and / or software configured to allocate resources 116 to consumers 112. Allocation refers to allocating resources 116 to requests of consumers 112 such that the resources allocated to a request service and / or process that request. Additionally or alternatively, allocation refers to designating resources 116 to consumers 112 such that resources designated for one consumer cannot service requests of other consumers.

[0024] In one embodiment, resource manager 110 performs allocation decisions based on consumer configurations 104 that include resource limits 106 and / or resource guarantees 108. Resource limits 106 are the maximum number of resources 116 that can be allocated to consumers 112 at any one time. Resource guarantees 108 are the minimum number of resources 116 guaranteed to consumers 112 at any one time. Example operations for enforcing resource limits 106 and / or resource guarantees 108 are further described below with reference to Figures 7-8.

[0025] Different consumer configurations 104 may correspond to different consumers 112. For example, one set of consumer configurations may correspond to consumer 112a, while another set of consumer configurations may correspond to consumer 112b.

[0026] In one or more embodiments, the data repository 102 is any type of storage unit and / or device for storing data (e.g., a file system, a database, a collection of tables, or any other storage mechanism). Furthermore, the data repository 102 may include multiple different storage units and / or devices. The multiple different storage units and / or devices may or may not be of the same type and may or may not be located at the same physical site. Furthermore, the data repository 102 may be implemented or executed on the same computing system as the resource manager 110. Alternatively or in addition, the data repository 102 may be implemented or executed on a computing system separate from the resource manager 110. The data repository 102 may be communicatively connected to the resource manager 110 via a direct connection or a network. The information describing the consumer configuration 104 may be implemented across any component in the system 100. However, this information is illustrated within the data repository 102 for purposes of clarity and explanation.

[0027] In one or more embodiments, CP solver 118 refers to hardware and / or software configured to determine a CP solution taking into account CP data model 232 and CP search directives 234. Further embodiments and / or examples of CP solver 118 are described below with reference to Figures 2A and 2B.

[0028] In one or more embodiments, system 100 of Figure 1 and system 200 of Figure 2A may be combined into a single system configured to manage the allocation of resources 116 to consumers 112. CP solver 118 of Figure 1 is equivalent to CP solver 218 of Figure 2A. Resource limits 106 and resource guarantees 108 of Figure 1 (used by resource manager 110) are equivalent to resource limits 206 and resource guarantees 208 of Figure 2A (output by CP solver 218), respectively.

[0029] 3. Constraint Programming Data Model and Search Instruction Generation System Architecture 2A and 2B illustrate a constraint programming data model and search command generation system according to one or more embodiments. As shown in FIG. 2A, the system 200 includes an excess distribution generator 222, a model and command generator 224, a data repository 202, a CP solver 218, a resource limit 206, and a resource guarantee 208. In one or more embodiments, the system 200 may include more or less components than those shown in FIG. 2A. The components shown in FIG. 2A may be local or remote to each other. The components shown in FIG. 2A may be implemented in software and / or hardware. Each component may be distributed to multiple applications and / or machines. Multiple components may be combined into one application and / or machine. Operations described with respect to one component may instead be performed by another component. As described above, the system 100 of FIG. 1 and the system 200 of FIG. 2A may be combined into a single system configured to allocate resources to consumers.

[0030] In one or more embodiments, the data repository 202 is any type of storage unit and / or device for storing data (e.g., a file system, a database, a collection of tables, or any other storage mechanism). Furthermore, the data repository 202 may include multiple different storage units and / or devices. The multiple different storage units and / or devices may or may not be of the same type and may or may not be located at the same physical site. Furthermore, the data repository 202 may be implemented or executed on the same computing system as the excess distribution generator 222, the model and command generator 224, and / or the CP solver 218. Alternatively or in addition, the data repository 202 may be implemented or executed on a computing system separate from the excess distribution generator 222, the model and command generator 224, and / or the CP solver 218. The data repository 202 may be communicatively coupled to the excess distribution generator 222, the model and command generator 224, and / or the CP solver 218 via a direct connection or a network.

[0031] The information describing the resource usage data set 226, the excess distribution 228, the vulnerability coefficients 230, the CP data model 232, and / or the CP search directives 234 may be implemented across any component in the system 200. However, this information is illustrated in the data repository 202 for purposes of clarity and explanation.

[0032] In one or more embodiments, the resource usage dataset 226 for a particular consumer includes data points that indicate the amount of resources used by a particular consumer over a period of time. The resource usage dataset 226 may be stored in the form of a log, a spreadsheet, and / or any other format. Resource usage may be measured in terms of number (such as the number of CPUs used), capacity (such as the capacity of memory space used), time (such as the time a communication channel is used), and / or other units. The resource usage dataset 226 may include data points for a particular time interval, which may be fixed and / or variable. Each data point is an actual measurement and / or an aggregated measurement. An actual measurement is the actual measured resource usage at a particular time. An aggregated measurement refers to the result of some operation (such as sum or average) on the actual measurements over a time window. As an example, a monitoring agent may detect the actual number of CPUs being used every second. The actual measurements of CPU usage starting at time=0 seconds and ending at time=9 seconds may be 2, 3, 4, 4, 4, 2, 3, 4, 5, 6. A monitoring agent may determine a running average of the amount of CPU being used every 3 seconds. At time=2, the average is calculated based on the actual measurements at time=0, time=1, and time=2. So starting at time=2 and ending at time=9 seconds, the average CPU usage might be 3.00, 3.67, 4.00, 3.33, 3.00, 3.00, 4.00, 5:00. The monitoring agent may report the average CPU usage as the "aggregated measurement."

[0033] In one or more embodiments, the excess distribution 228 indicates the number of data points that indicate a particular consumer's resource usage exceeds a critical value, the critical value ranging from zero to the total number of resources in the computing system. The number of such data points is based on the particular consumer's resource usage data set 226 over a particular historical period.

[0034] As an example, the resource usage data set may include the following data points: January 1, 2021, 10:00 AM: 2 CPU, January 1, 2021, 10:01am: 3CPU, January 1, 2021, 10:02 AM: 4 CPU, January 1, 2021, 10:03 AM: 4 CPUs, January 1, 2021, 10:04 AM: 4 CPU, January 1, 2021, 10:05 AM: 2 CPUs, January 1, 2021, 10:06 AM: 3 CPU, January 1, 2021, 10:07am: 4CPU, January 1, 2021, 10:08 AM: 5CPU, January 1, 2021, 10:09 AM: 6 CPU.

[0035] Therefore, the excess distribution of the resource usage dataset from 10:00am to 10:09am on January 1, 2021 would show the following: Critical value 0: 10 data points above the critical value; Critical value 1: 10 data points above the critical value; Critical value 2: 8 data points above the critical value; Critical value 3: 6 data points above the critical value; Critical value 4: 2 data points above the critical value; Critical value 5: 1 data point above the critical value; Critical value 6: 0 data points above the critical value.

[0036] The exceedance distribution will omit information greater than the minimum critical value associated with no exceedance (i.e., 6). Omission of such information is permissible because zero data points above the critical value of 6 necessarily implies zero data points above critical values ​​of 7, 8, 9, and 10. Additionally or alternatively, omission of such information may be desirable to ensure that resource guarantees are at or below the minimum critical value associated with no exceedance.

[0037] In one or more embodiments, the vulnerability factor 230 indicates a cost factor associated with a particular consumer vulnerability data point. For example, a consumer associated with a non-interruptible workload may be associated with a higher vulnerability factor than a consumer associated with an interruptible workload. A consumer associated with a higher priority (and / or a process having a higher priority) may be associated with a higher vulnerability factor than a consumer associated with a lower priority. A consumer associated with a higher revenue may be associated with a higher vulnerability factor than a consumer associated with a lower revenue. If the vulnerability factors 230 of all consumers are assumed to be the same, the vulnerability factor 230 of all consumers may be expressed as "1".

[0038] In one or more embodiments, the CP data model 232 refers to a particular organization, structure, and / or representation of information. The CP data model 232 declaratively expresses combinatorial properties of a problem in terms of constraints. The CP data model 232 may be implemented as a software data structure. The software data structure is machine readable. The CP data model 232 may include a set of data and / or instructions readable by one or more devices, including a hardware processor. The CP data model 232 may be used as an input parameter to hardware and / or software components, such as the CP solver 218.

[0039] 2B, the CP data model 232 includes a set of data model elements and a domain of possible values ​​that can be assigned to each element. A data model element cannot be assigned a value that is not within the data model element's domain. The data model elements may be implemented as arrays, vectors, linked lists, tables, software variables, constants, and / or other software data structures or data objects.

[0040] In one embodiment, the CP data model 232 includes several sets of data model elements. A set of limit elements 252 represent resource limitations for consumers. The domain of each limit element 252 ranges from zero to the total number of resources in the computing system.

[0041] A set of guarantee elements 254 represents resource guarantees to the consumer. The domain of each guarantee element 254 ranges from zero to the total number of resources in the computing system.

[0042] A set of vulnerability elements 256 represents a vulnerability count for a consumer based on resource usage over a particular historical period. The domain of each vulnerability element 256 ranges from zero to the total number of data points for the corresponding consumer within the particular historical period.

[0043] As an example, a computing system may include 8 CPUs and 4 database instances. The resource usage log for each database instance may show one data point per minute. The time period of interest may be from 10:00 AM to 10:09 AM on January 1, 2021. Thus, there are 10 data points for each consumer within the time period of interest.

[0044] Based on the above settings, the CP data model may have the following data model elements and corresponding domains: L[0]:{0, 1, 2, 3, 4, 5, 6, 7, 8} L[1]:{0, 1, 2, 3, 4, 5, 6, 7, 8} L[2]:{0, 1, 2, 3, 4, 5, 6, 7, 8} L[3]:{0, 1, 2, 3, 4, 5, 6, 7, 8} G[0]:{0, 1, 2, 3, 4, 5, 6, 7, 8} G[1]:{0, 1, 2, 3, 4, 5, 6, 7, 8} G[2]:{0, 1, 2, 3, 4, 5, 6, 7, 8} G[3]:{0, 1, 2, 3, 4, 5, 6, 7, 8} V[0]:{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10} V[1]:{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10} V[2]:{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10} V[3]:{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10} Let L[i] denote a restriction element, G[i] denote a guarantee element, and V[i] denote a vulnerability element, where i corresponds to a database instance.

[0045] Optionally, the CP data model 232 includes a set of per-consumer cost elements 258. The set of per-consumer cost elements 258 represents the respective costs of the consumers attributable to vulnerabilities over a particular historical period. The per-consumer cost element 258 for a particular consumer represents the product of (a) the vulnerability coefficient for the particular consumer and (b) the number of vulnerabilities for the particular consumer over the particular historical period. The domain of the per-consumer cost elements 258 for a particular consumer includes each product of (a) the vulnerability coefficient for the particular consumer and (b) each value in the domain of the vulnerability element for the particular consumer.

[0046] As an example, a computing system may include eight CPUs and four database instances. The resource usage log for each database instance may include one data point per minute. The time period of interest may be from 10:00 AM to 10:09 AM on January 1, 2021. The vulnerability factors for the four database instances may be 0.9, 0.6, 0.7, and 0.8, respectively.

[0047] As mentioned above, the CP data model may include the following vulnerability elements: V[0]:{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10} V[1]:{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10} V[2]:{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10} V[3]:{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10}.

[0048] The domain of the per-consumer cost elements of the first database instance CC[0] may include the products of 0.9 and each of the values ​​in the domain of V[0]. Thus, the CP data model may include the following per-consumer cost elements: CC[0]:{0.0, 0.9, 1.8, 3.6, 4.5, 5.4, 6.3, 7.2, 8.1, 9.0} CC[0]:{0.0, 0.6, 1.2, 1.8, 2.4, 3.0, 3.6, 4.2, 4.8, 5.4, 6.0} CC[0]:{0.0, 0.7, 1.4, 2.1, 2.8, 3.5, 4.2, 4.9, 5.6, 6.3, 7.0} CC[0]:{0.0, 0.8, 1.6, 2.4, 3.2, 4.0, 4.8, 5.6, 6.4, 7.2, 8.0}.

[0049] Optionally, CP data model 232 includes a total cost element 260. Total cost element 260 represents the total cost to consumers in a computing system due to vulnerabilities over a particular historical period. The domain of total cost element 260 indicates the possible total cost values ​​for consumers in a computing system. The minimum value in the domain is zero. The maximum value in the domain is the sum of the products of (a) each vulnerability coefficient for each consumer and (b) the total number of data points for each consumer in a particular historical period.

[0050] As an example, a computing system may include 8 CPUs and 4 database instances. The resource usage log for each database instance may show one data point per minute per consumer. The time period of interest may be from 10:00 AM to 10:09 AM on January 1, 2021. The vulnerability factors for the four database instances may be 0.9, 0.6, 0.7, and 0.8, respectively. Thus, the maximum value of the domain of total cost elements is 0.9×10+0.6×10+0.7×10+0.8×10=30.

[0051] In one embodiment, the CP data model 232 includes a number of constraints that define combinations of values ​​that can be assigned to a set of data model elements (such as the limit elements 252, guarantee elements 254, and vulnerability elements 256 described above).

[0052] A set of limit-guarantee constraints 262 relate link elements 252 and guarantee elements 254. The limit-guarantee constraints 262 require that the guarantee element 254 (for a particular consumer) is equal to the limit element 252 (for a particular consumer) divided by an oversubscription ratio, which is equal to the sum of the limit elements 252 divided by the total number of resources. The limit-guarantee constraints 262 can be expressed as follows:

[0053]

number

[0054] L[i] represents a limit element 252, G[i] represents a guarantee element 254, and L[c] represents a set of limit elements, where i corresponds to a particular consumer in the computing system.

[0055] Total resource constraint 264 relates guarantee factors 254 to the total number of resources in a computing system. Total resource constraint 264 requires that the sum of guarantee factors 254 is less than or equal to the total number of resources. Total resource constraint 264 can be expressed as follows:

[0056]

number

[0057] G[c] represents the set of guarantee elements 254 . A set of guarantee-vulnerability constraints 266 relate guarantee elements 254 and vulnerability elements 256. The guarantee-vulnerability constraints 266 require that the vulnerability element 256 (for a particular consumer) is equal to a vulnerability number that corresponds to a guarantee value that has been assigned (or should be assigned) to the guarantee element 254 (for a particular consumer) based on resource usage over a particular historical period.

[0058] In one embodiment, the guarantee-vulnerability constraint 266 is expressed as an element constraint. The element constraint accepts as inputs (a) an index element, and (b) a set of values. The index element input serves to index into the set of values ​​such that the element constraint returns a particular value from the set of values ​​that corresponds to a value that has been assigned (or should be assigned) to the index element. For the guarantee-vulnerability constraint 266, the index element is the guarantee element 254, and the set of values ​​is the excess distribution 228. Additionally, the index element outputs to a vulnerability element 256.

[0059] As an example, the CP data model may include the following data model elements: G[0]:{0, 1, 2, 3, 4, 5, 6, 7, 8} V[0]:{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10} Let G[0] denote the assurance factor corresponding to a particular consumer, and V[0] denote the vulnerability factor corresponding to the same particular consumer.

[0060] The excess distribution could be {10, 8, 7, 6, 4, 1, 1, 0}. The assurance-vulnerability constraints can be expressed as element constraints, with (a) G[0] as the input index element, (b) the excess distribution as the set of values ​​of the input, and (c) V[0] as the output element. Based on the element constraints, the value “0” assigned to G[0] is mapped to the particular set of values ​​“10”. Hence, the value “10” is returned. The value “1” assigned to G[0] is mapped to the particular set of values ​​“8”. Thus, the value “8” is returned. The value “2” assigned to G[0] is mapped to the particular set of values ​​“7”. Thus, the value “7” is returned. The returned values ​​must be within the domain of V[0].

[0061] Optionally, a set of per-consumer cost constraints 268 relate per-consumer vulnerability coefficients and vulnerability factors 256. Per-consumer cost constraints 268 (for a particular consumer) are the product of (a) the vulnerability coefficients (for the particular consumer) and (b) the vulnerability factors (for the particular consumer). Per-consumer cost constraints 268 can be expressed as follows: CC[i] = VF[i] × V[i] CC[i] represents the cost factor per consumer, VF[i] represents the vulnerability coefficient, and V[i] represents the vulnerability factor, where i corresponds to a consumer in the computing system.

[0062] Optionally, total cost constraint 270 relates vulnerability factors 256 to vulnerability coefficients across consumers in a computing system. Total cost constraint 270 is an inner product of a set of (a) vulnerability coefficients and (b) vulnerability factors. Total cost constraint 270 can be expressed as follows: TC = VF V TC represents the total cost factor, VF represents the consumer vulnerability factor, and V represents the vulnerability factor.

[0063] Alternatively, if a per-consumer cost component is used, the total cost constraint 270 can be expressed as:

[0064]

number

[0065] Let TC represent the total cost components and CC[c] represent the set of cost components per consumer.

[0066] Referring back to FIG. 2A , the CP search directives 234 are directives that guide the CP solver 218 in the process of determining an optimal CP solution given a particular CP data model 232. The CP search directives 234 guide the assignment of a set of values ​​to a set of data model elements that satisfy all constraints as specified by the CP data model 120. The CP search directives 234 prioritize the assignment of certain values ​​over other values ​​for one or more data model elements. Different CP search directives 234 for the same CP data model 232 may result in different CP solutions. Additionally or alternatively, different CP search directives 234 for the same CP data model 232 may result in different efficiency levels and / or execution times for obtaining a CP solution. The CP search directives 234 may be implemented as software data structures. The CP search directives 234 may include a set of computer-readable instructions. The CP search directives 234 may be used as input parameters to hardware and / or software components, such as the CP solver 218.

[0067] In one embodiment, the CP search directive 234 specifies an objective function, which is a minimization function associated with the number of vulnerabilities and / or the vulnerability cost of the computing system.

[0068] The CP search command 234 may include minimizing a maximum value of the vulnerability factors 256. Additionally or alternatively, the CP search command 234 may include minimizing a sum of the vulnerability factors 256. Additionally or alternatively, the CP search command 234 may include minimizing an average (e.g., mean, median, or mode) of the vulnerability factors 256.

[0069] If vulnerability costs are considered, the CP search directive 234 may include minimizing the maximum value of the cost elements 258 per consumer. Additionally, or alternatively, the CP search directive 234 may include minimizing the sum of the cost elements 258 per consumer (or the total cost element 260). Additionally, or alternatively, the CP search directive 234 may include minimizing the average (e.g., mean, median, or mode) of the cost elements 258 per consumer. Additional and / or alternative minimization functions may be used.

[0070] In one or more embodiments, model and command generator 224 refers to hardware and / or software configured to generate CP data model 232 and / or CP search command 234. Example operations for generating CP data model 232 are described below with reference to Figure 4. Example operations for generating CP search command 234 are described below with reference to Figures 5A-5C.

[0071] In one or more embodiments, the model and command generator 224 is implemented on one or more digital devices. The term "digital device" generally refers to any hardware device that includes a processor. A digital device may refer to a physical device that runs an application or a virtual machine. Examples of digital devices include computers, tablets, laptops, desktops, netbooks, servers, web servers, network policy servers, proxy servers, general purpose machines, function-specific hardware devices, mainframes, televisions, content receivers, set-top boxes, printers, mobile phones, smartphones, and personal digital assistants (PDAs).

[0072] In one or more embodiments, CP solver 218 refers to hardware and / or software configured to determine a CP solution using constraint programming techniques, given a CP data model 232 and CP search directives 234. The CP solution specifies resource limitations 206 and / or resource guarantees 208. Constraint programming techniques include backtracking algorithms, forward checking algorithms, constraint propagation, and the like. Example operations for applying the CP data model 232 and CP search directives 234 to the CP solver 218 are described below with reference to FIG. 6.

[0073] 4. Determining and Updating Resource Allocation Limits One or more of the operations illustrated in Figure 3 may be modified, rearranged, and / or omitted altogether, and thus the particular sequence of operations illustrated in Figure 3 should not be construed as limiting the scope of one or more embodiments.

[0074] FIG. 3 illustrates an example set of operations for updating resource limits and / or resource guarantees for consumers in a computing system based on resource usage over a particular historical period, according to one or more embodiments.

[0075] One or more embodiments include determining whether a trigger has occurred to update the resource limits and / or resource guarantees (act 302). A model and command generator (such as model and command generator 224 of FIG. 2A) determines whether a trigger has occurred to update the resource limits and / or resource guarantees. The trigger may be based on a schedule. As an example, the schedule may specify that the resource limits should be updated every three hours. Additionally or alternatively, the trigger may be based on the workload of one or more consumers in the computing system. As an example, the trigger may be that a consumer's workload exceeds a threshold. Another trigger may be that the aggregate workload of the consumers exceeds another threshold. Additional and / or alternative triggers may also be used.

[0076] One or more embodiments include determining consumers and resources in the computing system (act 304). A tracking system (which may be included in resource manager 110 of FIG. 1 and / or the model and command generator) tracks the consumers and resources in the computing system via a notification system, a heartbeat system, and / or other approaches. The tracking system stores identifiers of the consumers and resources in a data repository. The model and command generator retrieves identifiers of the consumers and resources in the computing system from the data repository.

[0077] One or more embodiments include obtaining a resource usage data set for the consumers over a particular time period (act 306). A monitoring system (which may be, but is not necessarily, included in the model and command generator) monitors resource usage for the consumers. The monitoring system detects, aggregates, measures, and / or otherwise obtains resource usage per consumer at regular time intervals (e.g., once per second). The monitoring system creates one data point for each resource usage measurement in each consumer's respective resource usage data set. The monitoring system stores the resource usage data set in a data repository. The model and command generator obtains the resource usage data set from the data repository.

[0078] In one embodiment, in each iteration to obtain updated resource limits and / or resource guarantees, the model and command generator uses resource usage data sets from the most recent historical period. Thus, the CP data models used to determine resource limits and / or resource guarantees are generated based on the most recent historical usage data. Thus, the resource limits and / or resource guarantees are adaptively updated based on changes in resource usage of consumers of the computing system.

[0079] As an example, the resource limits and resource guarantees of the computing system may be updated every hour. In a first iteration at 11:00 AM on January 1, 2021, the model and command generator may generate a CP data model based on the resource usage log from 10:00 AM to 10:59 AM on January 1, 2021. Thus, the resource limits and / or resource guarantees may be determined based on the resource usage data from 10:00 AM to 10:59 AM. In a second iteration at 12:00 PM on January 1, 2021, the model and command generator may generate a new CP data model based on the resource usage log from 11:00 AM to 11:59 AM on January 1, 2021. Thus, the resource limits and / or resource guarantees may be determined based on the resource usage data from 11:00 AM to 11:59 AM. Thus, the resource limits and / or resource guarantees may be continuously updated to be optimal based on the latest resource usage.

[0080] In one embodiment, at each iteration to obtain updated resource limits and / or resource guarantees, the model and command generator uses a prediction of resource usage for the next time period. The prediction of resource usage may be based on (a) a resource usage data set from a most recent historical time period, (b) a resource usage data set from a historical time period that shares characteristics with the next time period, and / or (c) other factors. The historical time period that shares characteristics with the next time period may be, for example, a historical time period that is associated with the same time of day, day of the week, and / or day of the month as the next time period. Thus, the resource limits and / or resource guarantees are adaptively updated based on the prediction of resource usage of consumers of the computing system.

[0081] One or more embodiments include determining a vulnerability coefficient for the consumer (act 308). A user and / or application specifies the vulnerability coefficient for the consumer. The vulnerability coefficient is stored in a data repository. The model and command generator retrieves the vulnerability coefficient from the data repository.

[0082] One or more embodiments include generating an excess distribution based on the resource usage dataset (operation 310). An excess distribution generator (which may be, but is not necessarily, included in the model and command generator, such as excess distribution generator 222 of FIG. 2A) generates the excess distribution based on the resource usage dataset.

[0083] First, the excess distribution generator determines the number of data points in the resource usage data set for a particular consumer in a particular historical period. The excess distribution generator continues the iterative process, with the number of data points in the particular historical period being the upper limit for the number of iterations. The excess distribution generator starts with a critical value of zero. The excess distribution generator counts the number of data points that show resource usage above the critical value. The excess distribution generator records the number in association with the critical value. The excess distribution increases the critical value and repeats the above process. The excess distribution generator repeats the above process until the critical value is greater than the number of data points in the particular historical period. Thus, the excess distribution generator obtains pairs of values, each pair including a critical value and the number of data points that show resource usage above the critical value. The pairs of values ​​form the excess distribution for the particular consumer. The excess distribution generator also repeats the above process for each consumer to obtain an excess distribution for each consumer.

[0084] The excess distribution generator stores the excess distributions of the consumers in the data repository. The model and command generator retrieves the excess distributions of the consumers from the data repository.

[0085] One or more embodiments include generating a CP data model (Act 312). The model and command generator generates the CP data model. An example operation for generating a CP data model is further described below with reference to FIG.

[0086] One or more embodiments include generating a CP search directive (act 314). The model and directive generator generates the CP search directive. Example operations for generating a CP search directive are further described below with reference to Figures 5A-5C. Each of Figures 5A-5C illustrates operations for generating a different alternative CP search directive. Another CP search directive not described with reference to Figures 5A-5C may be used instead.

[0087] One or more embodiments include inputting the CP data model and the CP search directives to a CP solver to obtain the resource limits and / or resource guarantees (act 316). The model and directives generator inputs the CP data model and the CP search directives to a CP solver (such as CP solver 218 of FIG. 2A). An example of an operation for applying the CP data model and the CP search directives to the CP solver is further described below with reference to FIG. 6.

[0088] One or more embodiments include applying the resource limits and / or resource guarantees to the computing system (act 318). The CP solver outputs the resource limits and / or resource guarantees to the consumers. A resource manager (such as resource manager 110 of FIG. 1) manages the computing system using the resource limits and / or resource guarantees. An example operation for assigning resources to consumers based on resource guarantees is described below with reference to FIG. 7. An example operation for queuing process requests that exceed resource limits is described below with reference to FIG. 8.

[0089] 5. Generation of constraint programming data model and constraint programming search instructions and application to constraint programming solver One or more of the operations illustrated in Figures 4-6 may be modified, rearranged, and / or omitted altogether, and therefore the particular sequence of operations illustrated in Figures 4-6 should not be construed as limiting the scope of one or more embodiments.

[0090] FIG. 4 illustrates an example set of operations for generating a constraint programming data model in accordance with one or more embodiments.

[0091] One or more embodiments include specifying a set of limit elements, each limit element representing a resource limitation for a respective consumer (act 402). A model and command generator (such as model and command generator 224 of FIG. 2A) specifies the set of limit elements. The set of limit elements may be represented as a vector, an array, and / or other data structure. Each limit element represents a resource limitation for a respective consumer in the computing system.

[0092] Additionally, the model and command generator specifies the domain of the constraint elements. Each domain may be represented as a vector, array, and / or other data structure. The domain of each constraint element ranges from zero to the total number of resources in the computing system.

[0093] One or more embodiments include specifying a set of guarantee elements, each guarantee element representing a resource guarantee for a respective consumer (act 404). The model and directive generator specifies the set of guarantee elements. The set of guarantee elements may be represented as vectors, arrays, and / or other data structures. Each guarantee element represents a resource guarantee for a respective consumer in the computing system.

[0094] Additionally, the model and directive generator specifies the domain of the guarantee elements. Each domain can be represented as a vector, array, and / or other data structure. The domain of each guarantee element ranges from zero to the total number of resources in the computing system.

[0095] One or more embodiments include specifying a set of vulnerability elements, where each vulnerability element represents a number of vulnerabilities for a respective consumer (act 406). The model and directive generator specifies the set of vulnerability elements. The set of vulnerability elements may be represented as a vector, an array, and / or other data structure. Each vulnerability element represents a number of vulnerabilities for a respective consumer in the computing system.

[0096] Additionally, the model and directive generator specifies a domain for the vulnerability elements. Each domain can be represented as a vector, array, and / or other data structure. The domain for each vulnerability element ranges from zero to the total number of data points for the corresponding consumer within a particular historical period.

[0097] In one embodiment, a user and / or application can specify a particular privilege consumer in a computing system. The user and / or application can specify a particular maximum value for the vulnerability count of the privilege consumer. The model and command generator can set the domain of the vulnerability element for the privilege consumer according to the specified maximum value. Additionally or alternatively, the user and / or application can also specify a particular minimum value for the resource guarantee of the privilege consumer. The model and command generator can set the domain of the guarantee element for the privilege consumer according to the specified minimum value.

[0098] As an example, a computing system may include 8 CPUs and 4 database instances. The resource usage log for each database instance may show one data point per minute. The time period of interest may be from 10:00 AM to 10:09 AM on January 1, 2021.

[0099] Without an externally specified maximum number of vulnerabilities, the set of vulnerability elements could be: V[0]:{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10} V[1]:{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10} V[2]:{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10} V[3]:{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10}.

[0100] However, if the user specifies that there cannot be any vulnerabilities in the first database, then the only number of vulnerabilities allowed in the first database is 0. All values ​​greater than 0 are removed from the domain of vulnerability elements in the first database. Thus, the set of vulnerability elements could be: V[0]: {0} V[1]:{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10} V[2]:{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10} V[3]:{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10}.

[0101] As another example, a computing system may include 8 CPUs and 4 database instances. The resource usage log for each database instance may show one data point per minute. The time period of interest may be from 10:00 AM to 10:09 AM on January 1, 2021.

[0102] In the absence of an externally specified minimum for resource guarantees, the set of guarantee elements could be: G[0]:{0, 1, 2, 3, 4, 5, 6, 7, 8} G[1]:{0, 1, 2, 3, 4, 5, 6, 7, 8} G[2]:{0, 1, 2, 3, 4, 5, 6, 7, 8} G[3]:{0, 1, 2, 3, 4, 5, 6, 7, 8}.

[0103] However, if the user specifies that the resource guarantee of the first database must be greater than or equal to 4, then 0, 1, 2, and 3 are not acceptable values ​​for the resource guarantee. Therefore, 0, 1, 2, and 3 are removed from the domain of the guarantee elements of the first database. Thus, the set of guarantee elements could be: G[0]: {4, 5, 6, 7, 8} G[1]:{0, 1, 2, 3, 4, 5, 6, 7, 8} G[2]:{0, 1, 2, 3, 4, 5, 6, 7, 8} G[3]:{0, 1, 2, 3, 4, 5, 6, 7, 8}.

[0104] As another example, the two examples above can be combined such that both the assurance and vulnerability elements are subject to change due to externally specified requirements. Combining the two examples above, the CP data model might contain the following assurance and vulnerability elements: G[0]: {4, 5, 6, 7, 8} G[1]:{0, 1, 2, 3, 4, 5, 6, 7, 8} G[2]:{0, 1, 2, 3, 4, 5, 6, 7, 8} G[3]:{0, 1, 2, 3, 4, 5, 6, 7, 8} V[0]: {0} V[1]:{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10} V[2]:{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10} V[3]:{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10}.

[0105] One or more embodiments include specifying a set of cost components per consumer and / or a total cost component (act 408).

[0106] Optionally, the model and directive generator specifies a set of per-consumer cost elements. The set of per-consumer cost elements may be represented as a vector, an array, and / or other data structure. Each per-consumer cost element represents a vulnerability cost for a respective consumer in the computing system. Additionally, the model and directive generator specifies a domain of the per-consumer cost elements. Each domain may be represented as a vector, an array, and / or other data structure. The domain of the per-consumer cost elements for a particular consumer includes each product of (a) the vulnerability coefficient for the particular consumer and (b) each value in the domain of the vulnerability elements for the particular consumer.

[0107] Optionally, the model and command generator specifies a total cost element. The total cost element represents a total vulnerability cost of the computing system. Additionally, the model and command generator specifies a domain for the total cost element. The domain may be represented as a vector, an array, and / or other data structure. The domain of the total cost element indicates possible total cost values ​​for consumers in the computing system. The minimum value in the domain is zero. The maximum value of the domain is the sum of the products of (a) each vulnerability coefficient for each consumer and (b) the total number of data points for each consumer in a particular historical period.

[0108] One or more embodiments include specifying a constraint relating a limit element and a guarantee element based on a total number of resources (act 410).

[0109] The model and command generator generates a set of limit-guarantee constraints 262, which require that the guarantee factor (for a particular consumer) is equal to the limit factor (for a particular consumer) divided by the oversubscription ratio, which is equal to the sum of the limit factors 252 divided by the total number of resources.

[0110] Additionally, the model and command generator generates a total resource constraint 264, which requires that the sum of the guarantee elements 254 be less than or equal to the total number of resources.

[0111] One or more embodiments include specifying a constraint requiring that each vulnerability element be assigned a vulnerability number that corresponds to the assurance value assigned to the assurance element (act 412). The model and directive generator generates a constraint requiring that each vulnerability element be assigned a vulnerability number that corresponds to the assurance value assigned to the assurance element based on resource usage over a particular historical period.

[0112] The model and command generator generates an element constraint. The model and command generator inputs the guarantee element for the particular consumer as an index element into the element constraint. The model and command generator inputs the excess distribution for the particular consumer (e.g., as determined in operation 310 of FIG. 3 ) as a set of values ​​into the element constraint. The model and command generator sets the vulnerability element for the particular consumer as an output element of the element constraint.

[0113] The model and directive generator repeats the above process for each consumer to generate a set of guarantee-vulnerability constraints.

[0114] One or more embodiments include specifying one or more constraints relating vulnerability coefficients and vulnerability elements for each consumer and / or across consumers (act 414).

[0115] Optionally, the model and command generator generates constraints relating the vulnerability coefficients and vulnerability factors for each consumer. The model and command generator sets a constraint requiring that the per-consumer vulnerability cost factor for a particular consumer is equal to the product of (a) the vulnerability coefficient for the particular consumer and (b) the vulnerability factor for the particular consumer. The model and command generator repeats the above process for each consumer to obtain a set of per-consumer cost constraints.

[0116] Optionally, the model and command generator generates constraints relating vulnerability coefficients and vulnerability elements across consumers of the computing system. The model and command generator sets a constraint requiring that the total vulnerability cost element be equal to the dot product of (a) the vulnerability coefficient for the consumer and (b) the set of vulnerability elements for the consumer. Alternatively, if per-consumer cost elements are used, the model and command generator sets a constraint requiring that the total vulnerability cost element be equal to the sum of the per-consumer cost elements for the consumer.

[0117] One or more embodiments include generating a CP data model including data model elements and constraints (act 416). The model and directive generator generates the CP data model. The CP data model includes a limit element, a guarantee element, a vulnerability element, and optionally a cost element per consumer and / or a total cost element. The CP data model further includes a limit-guarantee constraint, a total resource constraint, a guarantee-vulnerability constraint, and optionally a cost constraint per consumer and / or a total cost constraint.

[0118] The CP data model may be stored as a software data structure. The CP data model may include one or more arrays, vectors, linked lists, tables, software variables, constants, and / or data objects. The CP data model may include a set of data and / or instructions readable by one or more devices, including a hardware processor.

[0119] 5A-5C illustrate an example set of operations for generating constraint programming search commands in accordance with one or more embodiments, each of which illustrates an operation for generating a different alternative CP search command.

[0120] 5A, one or more embodiments include generating an objective function that minimizes a maximum value of a set of cost components per consumer (operation 502). A model and command generator (such as model and command generator 224 of FIG. 2A) generates an objective function that minimizes a maximum value of a set of cost components per consumer. The objective function can be expressed as:

[0121] min(max(CC)) CC represents a set of cost components per consumer.

[0122] One or more embodiments include generating a CP search directive including the objective function (operation 504). The model and directive generator generates the CP search directive. The CP search directive includes the objective function. The CP search directive may be stored as a software data structure. The CP search directive may include one or more arrays, vectors, linked lists, tables, software variables, constants, and / or data objects. The CP search directive may include a set of instructions executable by one or more devices including a hardware processor.

[0123] 5B, one or more embodiments include generating an objective function that minimizes a sum of a set of cost components per consumer (operation 506). The model and command generator generates an objective function that minimizes a sum of a set of cost components per consumer. The objective function can be expressed as:

[0124]

number

[0125] CC[i] represents the cost element per consumer, where i corresponds to a consumer in the computing system.

[0126] Alternatively, the objective function can be expressed as: min(TC) TC stands for total cost element.

[0127] One or more embodiments include generating a CP search directive that includes the objective function (ACT 508). Example operations for generating a CP search directive are described above with reference to ACT 504.

[0128] 5C, one or more embodiments include generating an objective function that minimizes an average of the set of cost components per consumer (operation 510). The model and command generator generates an objective function that minimizes an average of the set of cost components per consumer. The objective function can be expressed as:

[0129] min(ave(CC)) CC represents a set of cost components per consumer.

[0130] One or more embodiments include generating a CP search directive that includes the objective function (ACT 512). Example operations for generating a CP search directive are described above with reference to ACT 504.

[0131] 6 illustrates an example set of operations for applying a constraint programming data model and constraint programming search directives to a constraint programming solver in accordance with one or more embodiments. The operations in FIG. 6 assume a CP search directive that specifies minimizing the total cost of a computing system. However, modified operations can be used to implement different CP search directives.

[0132] One or more embodiments include accepting, by a CP solver, a CP data model and a CP search directive as input parameters (operation 602). A CP solver (such as CP solver 218 of FIG. 2A) accepts the CP data model and the CP search directive as input parameters. Example operations for generating a CP data model are described above with reference to FIG. 4. Example operations for generating a CP search directive are described above with reference to FIGs. 5A-5C.

[0133] One or more embodiments include determining whether the CP solver can return a CP solution based on the CP data model and the CP search directives (act 604). The CP solver applies one or more constraint programming techniques to the CP data model, guided by the CP search directives. The CP solver 132 determines a CP solution based on the CP data model and the CP search directives, or determines that no CP solution exists that satisfies all of the constraints of the CP data model.

[0134] Constraint programming techniques include, for example, constraint propagation, backtracking search algorithms, and / or forward checking algorithms. Constraint propagation involves removing inconsistent values ​​from the domain of data model elements in the CP data model. Backtracking search algorithms involve incrementally building up candidate CP solutions, such as abandoning a candidate as soon as the CP solver determines that it cannot complete a candidate that provides a valid CP solution. Forward checking algorithms include attempting to predict the impact of selecting one candidate over another for the CP solution, and determining the order in which to try candidate CP solutions. In one embodiment, the CP solver updates the domain of the data model elements as the CP solver traverses the data model elements and assigns values ​​to the data model elements. The CP solver removes values ​​from the domain that are no longer possible given any pre-assignments made so far. The CP solver removes values ​​from the domain that violate any constraints of the CP data model.

[0135] As an example, the CP solver may traverse each of a set of constraint elements specified by the CP data model to determine a CP solution. The CP solver may start with a particular constraint element. A value from the domain of the particular constraint element is pre-assigned to the particular constraint element. The CP solver then removes values ​​from the domains of other data model elements of the CP data model that are inconsistent with the pre-assigned value for the particular constraint element. If this pre-assignment violates a constraint specified by the CP data model (e.g., the domain of another data model element is completely removed), another value from the domain of the particular constraint element is pre-assigned to the particular constraint element. All values ​​from the domain of the particular constraint element are tried, and if they do not satisfy the constraint, a reassignment of one or more previously traversed constraint elements is performed. In particular, a value from the domain of a previously traversed constraint element was pre-assigned to the previously traversed constraint element. Based on the need for reassignment, another value from the domain of a previously traversed constraint element is pre-assigned to the previously traversed constraint element. The assignment of the particular constraint element is then retried.

[0136] Additionally, the CP solver may be guided by an objective function specified by a CP search directive. Based on the CP search directive, the CP solver may determine the order in which the assignment of values ​​to one or more constraint elements and / or guarantee elements is attempted. If the assignment of value A to a particular constraint element is attempted before value B and a valid CP solution is found, the assignment of value A is completed regardless of whether the assignment of value B reached a valid CP solution. Thus, the assignment of value A is prioritized over value B.

[0137] Thus, the CP solver may traverse each limiting element until each limiting element is assigned a limiting value from its respective domain without violating any constraints. The CP solver may determine the prior assignments as the final assignments. The CP solution may include final assignments of limit values ​​to limiting elements and / or guaranteed values ​​to guaranteed elements.

[0138] Alternatively, the CP solver may traverse each constraint element until an assignment of each constraint value to the constraint element is attempted, but no assignment satisfies the constraint. The CP solver may then determine that there is no valid CP solution for the CP data model. The CP solver may return a message indicating that no CP solution exists.

[0139] If no CP solution is returned, one or more embodiments include returning a resource limit and / or a resource guarantee using the last determined CP solution (operation 614). Since no CP solution is returned in the current iteration, the CP solver obtains the CP solution returned in the last iteration. The CP solver identifies the CP solution from the last iteration as the last determined CP solution. The CP solver also identifies the CP solution from the last iteration as the CP solution that satisfies the objective function of the CP search directive. If the objective function minimizes the total cost, the CP solution from the last iteration will be the CP solution associated with the lowest total cost (compared to other CP solutions that satisfy the constraints of the CP data model). The CP solution specifies a limit value assigned to a limit element and / or a guarantee value assigned to a guarantee element. The limit value is used as a resource limit for the consumer in the computing system. The guarantee value is used as a resource guarantee for the consumer in the computing system.

[0140] If there is no final CP solution determined (i.e., the CP solver is unable to return a CP solution in the first iteration of operation 602), then the CP solver generates a message indicating that no valid CP solution exists for the provided CP data model.

[0141] Conversely, if a CP solution is returned, one or more embodiments include determining whether an interrupt to the iterative process has been received (operation 606). As shown, operations 602-612 form an iterative process for finding a CP solution associated with the lowest total cost. A user and / or application may interrupt the iterative process. As an example, a user may indicate via a user interface that they would like the best CP solution determined so far, without waiting for the iterative process to complete.

[0142] If an interrupt is received, one or more embodiments include returning resource limits and / or resource guarantees using the last determined CP solution (operation 614). Because the CP solution is returned in the current iteration, the CP solver identifies the CP solution from the current iteration as the last determined CP solution. Example operations for determining resource limits and / or resource guarantees based on the CP solution are provided above in the description of operation 614.

[0143] In one embodiment, the CP solver may receive an interruption (at acts 602-604) while in the process of determining a CP solution. The interruption is received before the CP solution is returned in the current iteration. Thus, the CP solver identifies the CP solution from the last iteration as the CP solution that has been last determined.

[0144] The CP solution that is last determined based on the interruption is not necessarily the CP solution associated with the lowest total cost. A request to resume the iterative process for determining the CP solution associated with the lowest total cost may be received. In response to the request to resume, the iterative process may continue at operation 608.

[0145] If no interruption is received, one or more embodiments include identifying a total cost associated with the CP solution as the "current minimum total cost" (operation 608). The CP solver identifies a total cost assigned to the total cost elements associated with the CP solution obtained in operation 604 for the current iteration. The total cost is identified as the "current minimum total cost."

[0146] One or more embodiments include removing any values ​​from the domain of the total cost element that are greater than or equal to the current minimum total cost (operation 610). The CP solver uses the current minimum total cost as an upper bound on total cost values ​​that may be assigned to the total cost element during the next iteration. The CP solver removes values ​​from the domain of the total cost element that are greater than or equal to the current minimum total cost.

[0147] As an example, during the current iteration, the CP solver may return the current CP solution. The current CP solution may indicate that a total cost element is assigned a total cost value of "9." The CP solver may determine "9" as the current minimum total cost. The CP solver may determine that the current domain of the total cost element is {0, 4, 5, 8, 9, 11, 12}. The CP solver may remove total cost values ​​equal to or greater than 9 from the domain of the total cost element. Thus, the CP solver may change the domain of the total cost element to be {0, 4, 5, 8}.

[0148] One or more embodiments include modifying the CP data model (operation 612). The CP solver modifies the CP data model to include a total cost element whose domain has been reduced. Based on operation 610, the domain of the total cost element includes only total cost values ​​that are below the current minimum total cost.

[0149] The CP solver iterates through operations 602-612 with respect to the modified CP data model. In operation 602, the CP solver accepts the modified CP data model as an input parameter. In operations 604-608, the CP solver updates the current minimum total cost, assuming that the CP solver determines a new CP solution based on the modified CP data model. The CP solver identifies the total cost value assigned to the total cost element associated with the new CP solution as the current minimum total cost. In operation 610, the CP solver removes from the domain of the total cost element any values ​​equal to or greater than the current minimum total cost value. In operation 612, the CP solver again modifies the CP data model. The CP solver continues the iterative process until the CP solver is unable to find a CP solution in operation 604 or an interrupt is received in operation 606. As described above with reference to operation 614, if the CP solver is unable to find a CP solution in operation 604 or receives an interruption in operation 606, the CP solver returns the resource limitations and / or resource guarantees using the last determined CP solution. If the CP solver completes the iterations without interruption, the last determined CP solution becomes the optimal solution, i.e., there is no other solution with a lower cost.

[0150] In one or more embodiments, additional and / or alternative operations for applying the constraint programming data model and search directives to the constraint programming solver may be performed based on different search techniques. Examples of various search techniques are described below.

[0151] In one embodiment, the CP data model is modified before the first application of the CP solver in operation 602. A particular value within the range of possible total cost values ​​within the domain of the total cost element is determined as an initial cutoff value. The initial cutoff value may be, for example, the median of the possible total cost values. All values ​​above the initial cutoff value are removed from the domain of the total cost element. Thus, a first run of the CP solver is required to generate a CP solution with a total cost value below the initial cutoff value. If a valid CP solution exists, setting the initial cutoff value may allow the CP solver to more efficiently arrive at a CP solution with a lower total cost.

[0152] In one embodiment, the CP solver iterates until all possible CP solutions for a particular CP data model are found. The CP solver does not change the domain of the total cost element based on the previous iteration. Thus, the CP solver may find a CP solution with a higher total cost value than a previous CP solution. After determining all possible CP solutions, the CP solver compares the total cost values ​​associated with each possible CP solution. The CP solver then identifies the CP solution with the lowest total cost value.

[0153] In one embodiment, multiple CP solutions may be associated with the same lowest total cost. The iterative process may be modified to determine such a CP solution. At operation 610, remove all values ​​from the domain of the total cost element that are greater than the current minimum total cost (but keep all values ​​equal to the current minimum total cost). At operation 604, the CP solver attempts to find a CP solution that has not yet been found. If no further CP solutions are found, at operation 614, a set of last determined CP solutions with the same total cost are identified. A CP solution is selected from the set of last determined CP solutions with the same total cost. Any selection criteria may be used. As an example, the CP solver may select a CP solution from the set of last determined CP solutions with the same total cost that is associated with resource limitations that are most evenly distributed across the consumers.

[0154] In one or more embodiments, additional and / or alternative operations for applying the constraint programming data model and search directives to the constraint programming solver may be performed based on various CP search directives. Examples of various CP search directives are described below.

[0155] In one embodiment, the CP search directive specifies that the maximum value of the set of per-consumer cost components is to be minimized. Operations 602-606 and 612-614 remain the same. However, operation 608 is modified to identify the maximum value of the set of per-consumer cost components as the "current min-max per-consumer cost." Operation 610 is modified to remove any values ​​from the domain of each per-consumer cost component that are greater than or equal to the current min-max per-consumer cost.

[0156] In one embodiment, the CP search directive specifies minimizing the average of a set of cost elements per consumer. The CP data model includes an additional data model element called average cost element. The average cost element has a domain that includes the possible average costs per consumer in the computing system. The average cost element should equal the sum of the cost elements per consumer divided by the total number of consumers. Operations 602-606 and 612-614 remain the same. However, operation 608 is modified to identify the value of the average cost element as the "current minimum average cost." Operation 610 is modified to remove any value greater than or equal to the current minimum average cost from the domain of the average cost element.

[0157] 6. Enforcing resource guarantees and limits One or more of the operations illustrated in Figures 7-8 may be modified, rearranged, and / or omitted altogether, and therefore the particular sequence of operations illustrated in Figures 7-8 should not be construed as limiting the scope of one or more embodiments.

[0158] FIG. 7 illustrates an example set of operations for assigning resources to consumers based on resource guarantees that have been updated by a constraint programming solver in accordance with one or more embodiments.

[0159] One or more embodiments include obtaining resource limits and / or resource guarantees for consumers in a computing system (operation 702). A resource manager (such as resource manager 110 of FIG. 1) obtains resource limits and / or resource guarantees for consumers in the computing system based on a CP solution that has been determined by a CP solver (such as CP solver 218 of FIG. 2A). Example operations for determining a CP solution are described above with reference to FIG. 6, and in particular operation 614.

[0160] One or more embodiments include assigning resources to the consumer based on the resource guarantees (act 704). The resource manager assigns resources to the consumer based on the resource guarantees. In one embodiment, the resource manager assigns resources to the consumer based on the resource guarantees that have been determined by the CP solver without receiving user confirmation for the resource guarantees.

[0161] As an example, a computing system may include three database instances DB1, DB2, and DB3, and eight CPUs: CPU1, CPU2, CPU3, CPU4, CPU5, CPU6, CPU7, and CPU8. In a CP solution, the following resource guarantees may be specified: DB1 has 2 CPUs, 3 CPUs for DB2, For DB3, 1 CPU.

[0162] Thus, the resource manager can designate two CPUs to DB1, three CPUs to DB2, and one CPU to DB3. Specifically, the resource manager can designate CPU1 and CPU3 to DB1, CPU4, CPU6, and CPU8 to DB2, and CPU2 to DB3. The designated resources are reserved for use by the corresponding consumer and are not available to other consumers in the computing system. Thus, consumers using the designated resources are not vulnerable to interruptions caused by the requests of other consumers in the computing system. Meanwhile, CPU5 and CPU7 remain undesignated and are available for sharing between database instances.

[0163] One or more embodiments include determining whether the resource limits and / or resource guarantees have been updated (operation 706). The resource manager determines whether the resource limits and / or resource guarantees have been updated. If so, the resource manager redetermines the resource assignments in operation 704.

[0164] FIG. 8 illustrates an example set of operations for queuing process requests that exceed resource limits updated by a constraint programming solver in accordance with one or more embodiments.

[0165] One or more embodiments include determining that a particular consumer within the computing system has generated a new process request that requires a resource within the computing system (act 802). A resource manager (such as resource manager 110 of FIG. 1) determines that a particular consumer within the computing system has generated a new process request that requires a resource within the computing system. The particular consumer may send the new process request to the resource manager for scheduling. Additionally or alternatively, the particular consumer may send a notification message to the resource manager to inform the resource manager about the new process request.

[0166] One or more embodiments include determining a number of pending process requests made by the particular consumer (act 804). The resource manager determines the number of process requests (including new process requests) that (a) have been made by the particular consumer and (b) have not yet been fully processed.

[0167] One or more embodiments include determining whether the number of pending process requests is less than or equal to the resource guarantee for the particular consumer (operation 806). The resource manager determines whether the number of pending process requests is less than or equal to the resource guarantee for the particular consumer. In one embodiment, the resource manager uses the resource guarantee determined by the CP solver as a threshold without receiving user confirmation of the resource guarantee determined by the CP solver.

[0168] If the number of pending process requests is less than or equal to the resource guarantee, one or more embodiments include allocating the specified resource to the new process request (operation 808). The resource manager allocates the resource to the new process request. In one embodiment, the resource manager is set to allocate the specified resource as much as possible. Because the number of pending process requests is less than or equal to the resource guarantee, at least one of the specified resources is available. Thus, the resource manager allocates an available dedicated resource to the new process request.

[0169] If the number of pending process requests exceeds the resource guarantee, one or more embodiments include determining whether the number of pending process requests is greater than a resource limit for the particular consumer (act 810). The resource manager determines whether the number of pending process requests exceeds a resource limit for the particular consumer. In one embodiment, the resource manager uses the resource limit determined by the CP solver as a threshold without receiving user confirmation of the resource limit determined by the CP solver.

[0170] If the number of pending process requests is greater than the resource limit, one or more embodiments include queuing the new process request (act 814). The resource manager queues the new process request. Resources within the computing system are not allocated to the process request in the queue until the process request is removed from the queue.

[0171] If the number of process requests of a particular consumer being processed falls below the resource limit, the resource manager has the option to remove the process request from the queue of the particular consumer. If another resource is available, the resource manager may allocate the available resource to the process request being removed from the queue. If no resource is available, the resource manager may optionally suspend the ongoing execution of the particular resource of a process request made by another consumer. The resource manager allocates the particular resource to the process request being removed from the queue of the particular consumer. Alternatively, if no resource is available, the resource manager may refrain from suspending any ongoing execution. Whether to suspend may depend on various factors, examples of which are described below with reference to operation 812.

[0172] If the number of pending process requests is (a) greater than the resource guarantee and (b) less than or equal to the resource limit, one or more embodiments include either allocating resources to the new process request or queuing the process request (operation 812). If at least one resource is available and the queue for the particular consumer is empty, the resource manager allocates the available resources to the new process request. If at least one resource is available and the queue for the particular consumer is not empty, the resource manager removes the process request from the head of the queue and allocates the available resources to the process request. The resource manager adds the new process request to the end of the queue.

[0173] If no resources are available, the resource manager may evaluate various factors to determine whether to suspend ongoing execution of a process request made by another consumer. Examples of factors include priority levels associated with the new process request made by the particular consumer and other process requests that are running, priority levels associated with the particular consumer and other consumers, the time that the other process requests are running, the predicted time remaining until the execution of the other process requests is completed, and the predicted time until the execution of the new process request made by the particular consumer is completed. Based on the various factors, the resource manager may suspend ongoing execution of a process request by the particular resource. The resource manager may allocate the particular resource to a process request that is being removed from the queue of the particular consumer (or to a new process request being created by the particular consumer if the queue is empty). Alternatively, based on the various factors, the resource manager may queue the new process request and continue to wait before allocating resources to any process request made by the particular consumer.

[0174] 7. Hardware Overview According to one embodiment, the techniques described herein are implemented by one or more special-purpose computing devices. The special-purpose computing devices may include digital electronic devices such as one or more application-specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), or network processing units (NPUs) that may be hardwired to perform the techniques or permanently programmed to perform the techniques, or may include one or more general-purpose hardware processors that are programmed to perform the techniques according to program instructions in firmware, memory, other storage, or combinations thereof. Such special-purpose computing devices may also combine custom hardwired logic, ASICs, FPGAs, or NPUs with custom programming to realize the techniques. The special-purpose computing devices may be desktop computer systems, portable computer systems, handheld devices, network devices, or other devices that incorporate hardwired and / or program logic to implement the techniques.

[0175] For example, Figure 9 is a block diagram illustrating a computer system 900 in which embodiments of the invention may be implemented. Computer system 900 includes a bus 902 or other communication mechanism for communicating information, and a hardware processor 904 coupled to bus 902 for processing information. Hardware processor 904 may be, for example, a general-purpose microprocessor.

[0176] Computer system 900 also includes a main memory 906, such as a random access memory (RAM) or other dynamic storage device, coupled to bus 902 for storing information and instructions executed by processor 904. Main memory 906 may also be used for storing temporary variables or other intermediate information during execution of instructions executed by processor 904. Such instructions, when stored in a non-transitory storage medium accessible to processor 904, render computer system 900 into a special-purpose machine that is customized to perform the operations specified in the instructions.

[0177] Computer system 900 further includes a read only memory (ROM) 908 or other static storage device coupled to bus 902 for storing static information and instructions for processor 904. A storage device 910, such as a magnetic disk or optical disk, is provided and coupled to bus 902 for storing information and instructions.

[0178] The computer system 900 may be coupled via bus 902 to a display 912, such as a cathode ray tube (CRT), for displaying information to a computer user. An input device 914, including alphanumeric and other keys, is coupled to bus 902 for communicating information and command selections to the processor 904. Another type of user input device is a cursor control 916, such as a mouse, trackball, or cursor direction keys, for communicating directional information and command selections to the processor 904 and for controlling cursor movement on the display 912. This input device typically has two degrees of freedom in two axes, a first axis (e.g., x) and a second axis (e.g., y), allowing the device to specify a position in a plane.

[0179] The computer system 900 can implement the techniques described herein using customized hardwired logic, one or more ASICs or FPGAs, firmware and / or program logic that, in combination with the computer system, makes or programs the computer system 900 into a special-purpose machine. According to one embodiment, the techniques herein are performed by the computer system 900 in response to the processor 904 executing one or more sequences of one or more instructions contained in the main memory 906. Such instructions may be read into the main memory 906 from another storage medium, such as the storage device 910. Execution of the sequences of instructions contained in the main memory 906 causes the processor 904 to perform the process steps described herein. In alternative embodiments, hardwired circuitry may be used in place of or in combination with software instructions.

[0180] The term "storage medium" as used herein refers to any non-transitory medium that stores data and / or instructions that cause a machine to operate in a specific manner. Such storage media can include non-volatile media and / or volatile media. Non-volatile media include, for example, optical or magnetic disks, such as storage device 910. Volatile media include dynamic memory, such as main memory 906. Common forms of storage media include, for example, floppy disks, flexible disks, hard disks, solid state drives, magnetic tape or other magnetic data storage media, CD-ROMs, other optical data storage media, any physical media with a pattern of holes, RAM, PROMs, and EPROMs, FLASH-EPROMs, NVRAMs, other memory chips or cartridges, Content Addressable Memory (CAM), and Ternary Content Addressable Memory (TCAM).

[0181] Storage media is distinct from but may be used in combination with transmission media. Transmission media involves transferring information between storage media. For example, transmission media include coaxial cables, copper wire and fiber optics, including the wires that comprise bus 902. Transmission media can also take the form of acoustic or light waves, such as those generated during radio wave or infrared data communications.

[0182] Various forms of media may be involved in carrying one or more sequences of one or more instructions to the processor 904 for execution. For example, the instructions may initially be stored on a magnetic disk or solid state drive of a remote computer. The remote computer can load the instructions into a dynamic memory and send the instructions over a telephone line using a modem. A modem local to the computer system 900 can receive the data on the telephone line and use an infrared transmitter to convert the data to an infrared signal. An infrared detector can receive the data carried in the infrared signal and appropriate circuitry can place the data on the bus 902. The bus 902 carries the data to the main memory 906, from which the processor 904 retrieves and executes the instructions. The instructions received by the main memory 906 may optionally be stored on storage device 910 either before or after execution by the processor 904.

[0183] Computer system 900 also includes a communication interface 918 coupled to bus 902. The communication interface 918 provides a two-way data communication coupling to a network link 920 that is connected to a local network 922. For example, the communication interface 918 may be an Integrated Services Digital Network (ISDN) card, a cable modem, satellite modem, or a modem that provides a data communication connection to a corresponding type of telephone line. As another example, the communication interface 918 may be a local area network (LAN) card that provides a data communication connection to a compatible LAN. Wireless links may also be implemented. In such implementations, the communication interface 918 sends and receives electrical, electromagnetic or optical signals that carry digital data streams representing various types of information.

[0184] Network link 920 typically provides data communication through one or more networks to other data devices. For example, network link 920 may provide a connection through a local network 922 to a host computer 924 or to data equipment operated by an Internet Service Provider (ISP) 926. The ISP 926 in turn provides data communication services through a worldwide packet data communication network now commonly referred to as the "Internet" 928. Local network 922 and Internet 928 both use electrical, electromagnetic or optical signals that carry digital data streams. The signals through the various networks, the signals on network link 920 and through communication interface 918, which carry the digital data to and from computer system 900, are exemplary forms of transmission media.

[0185] Computer system 900 can send messages and receive data, including program code, through the network(s), network link 920 and communication interface 918. In the Internet example, a server 930 might transmit a requested code for an application program through Internet 928, ISP 926, local network 922 and communication interface 918.

[0186] The received code may be executed by processor 904 as it is received, and / or stored in storage device 910, or other non-volatile storage for later execution.

[0187] 8.Others;Extensions Embodiments are directed to systems with one or more devices that include a hardware processor and are configured to perform any of the operations described herein and / or recited in any of the claims below.

[0188] In one embodiment, a non-transitory computer-readable storage medium comprises instructions that, when executed by one or more hardware processors, cause performance of any of the operations described and / or claimed herein.

[0189] Any combination of the features and functions described herein may be used according to one or more embodiments. In the foregoing specification, embodiments have been described with reference to many specific details that may vary from implementation to implementation. Therefore, the specification and drawings should be regarded in an illustrative sense, not a restrictive sense. The sole and exclusive indication of the scope of the invention, and what the applicant intends to be the scope of the invention, is the literal equivalent range of the set of claims issued from this application in the particular form in which such claims are issued, including any subsequent amendments.

Claims

1. A program for causing a processor to execute a method, comprising: The method comprises: determining a set of consumers and a set of resources in a computing system; obtaining a set of resource usage data sets corresponding respectively to said set of consumers over a particular time period; A particular resource usage dataset corresponds to a particular consumer, The method further includes generating a set of excess distributions each corresponding to the set of consumers; Generating a particular excess distribution corresponding to the particular consumer includes: determining a respective number of data points in the particular resource usage data set that indicate resource usage by the particular consumer exceeds each of a set of critical values; The method further includes generating a constraint programming data model; Generating a constraint programming data model includes: specifying a set of limit elements, each limit element representing a resource limit for said set of consumers; a resource limit for the particular consumer is an upper limit on the number of resources that the particular consumer is allowed to use at a particular time; specifying a set of guarantee elements, each of which represents a resource guarantee for said set of consumers; a resource guarantee for the particular consumer that is a lower bound on the number of resources that the particular consumer may access at a particular time; a particular assurance element of said set of assurance elements corresponds to said particular consumer; determining a set of vulnerability elements based on said set of resource usage data sets, each of said vulnerability elements representing a vulnerability number of said set of consumers; the vulnerability count for the particular consumer is a number of data points that indicate a vulnerability based on a resource usage data set over the particular time period for the particular consumer; a particular vulnerability element of said set of vulnerability elements corresponds to said particular consumer; specifying a set of assurance-vulnerability constraints associated with said set of assurance elements and said set of vulnerability elements; the guarantee-vulnerability constraint corresponding to the particular consumer requires assigning a counter value to the particular vulnerability element that corresponds to a guarantee value assigned to the particular guarantee element based on the particular excess distribution; The method further includes generating constraint programming search commands; Generating a constraint programming search directive comprises: specifying a minimization objective function associated with said set of vulnerability elements; The method further includes applying the constraint programming data model and the constraint programming search directives to a constraint programming solver to obtain at least one of the resource limitations for the set of consumers and the resource guarantees for the set of consumers. program.

2. a particular limiting element of the set of limiting elements corresponds to the particular consumer; Generating a constraint programming data model includes: specifying a set of limit-guarantee constraints associated with the set of limit elements and the set of guarantee elements; a limit-guarantee constraint corresponding to the particular consumer requires that the particular guarantee factor is equal to the particular limit factor divided by an oversubscription ratio; The program of claim 1 , wherein the oversubscription ratio is equal to the sum of the set of limiting factors divided by the total number of resources.

3. a particular limiting element of the set of limiting elements corresponds to the particular consumer; Generating a constraint programming data model includes: specifying a total resource constraint and a total number of the resources associated with the set of guarantee elements; The program of claim 1 , wherein the total resource constraint requires that the sum of the set of guarantee elements be less than or equal to the total number of resources.

4. The minimization objective function is: the maximum value of said set of vulnerability elements; the sum of said set of vulnerability elements; and The average of said set of vulnerability elements The program according to claim 1 , further comprising:

5. The method further comprises: obtaining a set of vulnerability coefficients respectively corresponding to said set of consumers; a particular vulnerability factor corresponds to said particular consumer; Generating a constraint programming data model includes: specifying a set of per-consumer cost elements each representing a cost to the consumer attributable to a vulnerability over the particular time period, a particular per-consumer cost element corresponding to the particular consumer; specifying a set of per-consumer cost constraints associated with said set of vulnerability coefficients and said set of vulnerability factors, wherein a per-consumer cost constraint corresponding to said particular consumer requires that said particular per-consumer cost factor be equal to the product of (a) said particular vulnerability coefficient and (b) said particular vulnerability factor; The program of claim 1 , wherein the minimization objective function is further associated with the set of per-consumer cost components.

6. The minimization objective function is: the maximum value of said set of cost elements per consumer; the sum of said set of cost elements per consumer; and The average of the above set of cost elements per consumer The program according to claim 5 , further comprising:

7. The method further comprises: obtaining a set of vulnerability coefficients respectively corresponding to said set of consumers; a particular vulnerability factor corresponds to said particular consumer; Generating a constraint programming data model includes: specifying a total cost component representing a total cost to the consumer attributable to vulnerabilities over the specified time period; specifying a total cost constraint associated with said set of vulnerability coefficients and said set of vulnerability elements, said total cost constraint requiring said total cost element to be equal to a dot product of (a) said set of vulnerability coefficients and (b) said set of vulnerability elements; The program of claim 1 , wherein the minimization objective function is further associated with the total cost component.

8. The method further comprises: obtaining a set of vulnerability coefficients respectively corresponding to said set of consumers; a particular vulnerability factor corresponds to said particular consumer; Generating a constraint programming data model includes: specifying a set of per-consumer cost elements each representing a cost to the consumer attributable to a vulnerability over the particular time period, a particular per-consumer cost element corresponding to the particular consumer; specifying a total cost component representing a total cost to the consumer attributable to the vulnerability over the particular time period; specifying a set of per-consumer cost constraints associated with said set of vulnerability coefficients and said set of vulnerability factors, wherein a per-consumer cost constraint corresponding to said particular consumer requires that said particular per-consumer cost factor be equal to the product of (a) said particular vulnerability coefficient and (b) said particular vulnerability factor; specifying a total cost constraint associated with said set of vulnerability coefficients and said set of vulnerability elements, said total cost constraint requiring said total cost element to be equal to a sum of said set of per-consumer cost elements; The program of claim 1 , wherein the minimization objective function is further associated with at least one of: (a) the set of per-consumer cost components; and (b) the total cost component.

9. The assurance-vulnerability constraints are expressed as element constraints, The particular guarantee element is used as an input index element to the element constraint; The particular excess distribution is used as the input set of values ​​to the element constraints; and The program of claim 1 , wherein the particular vulnerability element is used as an output to the element constraint.

10. The program of claim 1 , wherein the set of resource usage data over a particular period of time comprises resource usage data measured over a historical period of time.

11. The program of claim 1 , wherein the set of resource usage data over a particular time period includes projected resource usage data over a future time period.

12. The method further comprising:

2. The program of claim 1, further comprising assigning to the particular consumer a number of the set of resources equal to a resource guarantee determined for the particular consumer by the constraint programming solver.

13. 13. The program of claim 12, wherein resources designated for the particular consumer cannot be used by other consumers in the set of consumers.

14. The method further comprising: determining whether a number of pending process requests made by the particular consumer is greater than a resource limit determined for the particular consumer by the constraint programming solver; 2. The program of claim 1, further comprising, in response to determining that a number of pending process requests exceeds the resource limit, adding a new process request made by the particular consumer to a queue rather than allocating resources to the new process.

15. 15. The program of claim 14, wherein no resources are allocated to any process request in the queue.

16. The method further comprising: obtaining a second set of resource usage data sets corresponding respectively to said set of consumers over a second time period; generating a second set of excess distributions respectively corresponding to said set of consumers based on said second set of resource usage data sets; generating a second constraint programming data model based on the second set of excess distributions; and 2. The program of claim 1, further comprising applying at least the second constraint programming data model to the constraint programming solver to obtain at least one of updated resource limitations for the set of consumers and updated resource guarantees for the set of consumers.

17. 2. The program of claim 1, wherein the constraint programming solver periodically updates at least one of the resource limits for the set of consumers and the resource guarantees for the set of consumers based on an updated resource usage data set.

18. The method further comprises: obtaining a set of vulnerability coefficients respectively corresponding to said set of consumers; a particular vulnerability factor corresponds to said particular consumer; a particular limiting element of the set of limiting elements corresponds to the particular consumer; Generating a constraint programming data model includes: specifying a set of limit-guarantee constraints associated with the set of limit elements and the set of guarantee elements; a limit-guarantee constraint corresponding to the particular consumer requires that the particular guarantee factor is equal to the particular limit factor divided by an oversubscription ratio; the oversubscription ratio is equal to the sum of the set of limiting factors divided by the total number of resources; specifying a total resource constraint and a total number of the resources associated with the set of guarantee elements; the total resource constraint requires that the sum of the set of guarantee elements is less than or equal to the total number of resources; specifying a set of per-consumer cost elements each representing a cost to the consumer attributable to a vulnerability over the particular time period, a particular per-consumer cost element corresponding to the particular consumer; specifying a total cost component representing a total cost to the consumer attributable to the vulnerability over the particular time period; specifying a set of per-consumer cost constraints associated with said set of vulnerability coefficients and said set of vulnerability factors, wherein a per-consumer cost constraint corresponding to said particular consumer requires that said particular per-consumer cost factor be equal to the product of (a) said particular vulnerability coefficient and (b) said particular vulnerability factor; and specifying a total cost constraint associated with the set of vulnerability coefficients and the set of vulnerability elements, the total cost constraint comprising: (i) said total cost component is equal to the dot product of (a) said set of vulnerability coefficients and (b) said set of vulnerability components; and (ii) the total cost element is equal to the sum of the set of cost elements per consumer; At least one of the following is required: The minimization objective function is: the maximum value of said set of vulnerability elements; the sum of said set of vulnerability elements; and an average of said set of vulnerability elements; the maximum value of said set of cost elements per consumer; a sum of said set of cost elements per consumer; the average of said set of cost elements per consumer; and said total cost elements, Minimize at least one of The assurance-vulnerability constraints are expressed as element constraints, The particular guarantee element is used as an input index element to the element constraint; The particular excess distribution is used as an input set of values ​​to the element constraints; and The identified vulnerability elements are used as an output to the element constraints; the set of resource usage data sets over a particular time period includes at least one of: (a) resource usage data that is measured over a historical time period; and (b) resource usage data that is projected over a future time period; The method further includes assigning to the particular consumer a number of the set of resources equal to a resource guarantee determined for the particular consumer by the constraint programming solver; resources designated for said particular consumer cannot be used by other consumers in said set of consumers; determining whether a number of pending process requests made by the particular consumer is greater than a resource limit determined for the particular consumer by the constraint programming solver; in response to determining that the number of pending process requests exceeds the resource limit, adding new process requests made by the particular consumer to a queue rather than allocating resources to the new process; no resources are allocated to any process request in said queue; obtaining a second set of resource usage data sets corresponding respectively to said set of consumers over a second time period; generating a second set of excess distributions respectively corresponding to said set of consumers based on said second set of resource usage data sets; generating a second constraint programming data model based on the second set of excess distributions; and applying at least the second constraint programming data model to the constraint programming solver to obtain at least one of updated resource limitations for the set of consumers and updated resource guarantees for the set of consumers; 2. The program of claim 1, wherein the constraint programming solver periodically updates at least one of the resource limits for the set of consumers and the resource guarantees for the set of consumers based on an updated resource usage data set.

19. A method according to any one of claims 1 to 18.

20. A system comprising means for executing a method according to any one of claims 1 to 18.

Images