Computing load balancing in distributed environment
By analyzing the load and resources of the computing device in the cloud computing system, and using the assigned solver to transfer projects between multiple computing devices, the problem of imbalance in the computing device utilization is solved, load and cost optimization is achieved, and the system is improved Efficiency and user experience.
Patent Information
- Application Number
- CN202380026627.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2022-03-10
- Filing Date
- 2023-03-08
- Publication Date
- 2024-11-08
AI Technical Summary
The imbalance in the calculation utilization rate between computing devices in cloud computing systems leads to over-utilization of some computing devices while insufficient utilization of other computing devices, affecting system efficiency.
By assigning a solver to analyze the computing load and available resources of the computing device, determine the calculation utilization rate, and transfer items between multiple sets of computing devices to reduce the calculation utilization difference, using naive solvers, greedy algorithms or mixed integer problem solvers to optimize storage and communication costs.
Load balancing between computing devices is realized, computing differences are reduced, storage and communication costs are optimized, and system efficiency and user experience are improved.
Smart Images

Figure CN118922819A_ABST
Abstract
Description
[0001] CROSS-REFERENCE TO RELATED APPLICATIONS
[0002] This application is a continuation of U.S. patent application No. 17 / 691,570 filed on March 10, 2022, the disclosure of which is hereby incorporated by reference herein. Background Art
[0003] Cloud computing systems sometimes include a distributed data analysis engine that operates in multiple data centers distributed around the world. Each data center contains one or more servers. Users of such cloud computing systems can create organizations and projects. Within a project, the distributed data analysis engine allows users to create data sets and tables. Internally, tables are divided into data replication units, called storage sets. Each storage set corresponds to one or more files stored on a server. Although users typically query their own data sets, a user can also share a data set with another user or make a data set publicly available to many users. Multiple data sets can be linked together when querying, which potentially requires the system to read data from a large number of different data sets that may belong to any user.
[0004] Data sets can be strategically stored across a distributed computing system. Typically, techniques for strategic data storage involve identifying items with shared data sets and storing items with the most frequently shared data sets on the same computing device within the distributed computing system.
[0005] However, policy data storage can result in a large group of items being gathered at a single computing device, while few items are left on other computing devices. This results in an imbalance in computing utilization between computing devices, with some computing devices being over-utilized and others being under-utilized. Summary of the invention
[0006] The present disclosure calculates the utilization difference between the most overutilized computing device and the most underutilized computing device of the computing system, and then determines the rebalancing of the computing load to reduce the utilization difference. The assignment solver can analyze the computing utilization as well as other factors. The project assignment set that minimizes the computing utilization difference between the computing devices of the system can be an optimized storage solution.
[0007] One aspect of the present disclosure relates to a method for load balancing between multiple groups of computing devices, each group including one or more corresponding computing devices. The method may include: determining, by one or more processors, a first group of computing devices with the highest computing utilization level and a second group of computing devices with the lowest computing utilization level; determining, by one or more processors, a transfer of one or more items between the multiple groups of computing devices to reduce the computing utilization difference between the first group of computing devices and the second group of computing devices; and instructing, by one or more processors, the multiple groups of computing devices to perform the determined transfer.
[0008] In some examples, determining the computing utilization density may include: for each group of computing devices: determining, by one or more processors, a computing load of the group of computing devices; determining, by one or more processors, an amount of available computing resources for the group of computing devices; and calculating, by one or more processors, computing utilization of the group of computing devices based on the computing load and the amount of available computing resources; and determining, by one or more processors, a first group of computing devices and a second group of computing devices based on the calculated computing utilization.
[0009] In some examples, the computational load may be the amount of computation performed within a predefined duration, the amount of available computing resources may be the maximum amount of computation that the group of computing devices can perform within the predefined duration, and the computational utilization may be the ratio of the computational load to the amount of available computing resources.
[0010] In some examples, the predefined duration may be a day, a week, a month, or a year.
[0011] In some examples, determining the computational load may include: determining, by one or more processors, multiple projects assigned to the group of computing devices; for each determined project: accessing, by one or more processors, a job log of queries for the project; determining, by one or more processors, a project computational load for the project based on the job log of queries; and calculating, by one or more processors, a sum of the project computational loads for the multiple projects.
[0012] In some examples, determining a transfer of one or more items between multiple groups of computing devices to reduce a computing utilization difference between a first group of computing devices and a second group of computing devices may include at least one of: determining that the transfer reduces computing utilization of the first group of computing devices; or determining that the transfer increases computing utilization of the second group of computing devices.
[0013] In some examples, the method may also include storing, by the one or more processors, an assignment map indicating a mapping of items to the multiple groups of computing devices based on the determined transfer of the one or more items between the multiple groups of computing devices.
[0014] In some examples, the method may also include: determining, by one or more processors, one or more secondary cost factors associated with transferring one or more items between multiple groups of computing devices; assigning, by one or more processors, a first weighted value to a computing utilization difference and assigning a corresponding weighted value to each of the one or more secondary factors; and calculating, by the one or more processors, a sum of the weighted computing utilization difference and the weighted secondary cost factors. The determined transfer of the one or more items between the multiple groups of computing devices may reduce the calculated sum.
[0015] In some examples, the method may also include that one or more secondary cost factors may include at least one of: network bandwidth costs for transferring one or more items between multiple groups of computing devices; or network communication costs for transmitting data between multiple groups of computing devices.
[0016] In some examples, determining the transfer of one or more items between multiple groups of computing devices may be performed by one of a naive solver, a greedy algorithm solver, or a mixed integer problem solver.
[0017] Another aspect of the present disclosure relates to a system for load balancing between multiple groups of computing devices, each group including one or more corresponding computing devices. The system may include: one or more memories; and one or more processors, the one or more processors communicating with the one or more memories. The one or more processors may be configured to: determine a first group of computing devices with the highest computing utilization level and a second group of computing devices with the lowest computing utilization level; determine a transfer of one or more items between multiple groups of computing devices to reduce the difference in computing utilization between the first group of computing devices and the second group of computing devices; and instruct multiple groups of computing devices to perform the determined transfer.
[0018] In some examples, one or more processors may be configured to: for each group of computing devices: determine the computing load of the group of computing devices; determine the amount of available computing resources for the group of computing devices; and calculate computing utilization of the group of computing devices based on the computing load and the amount of available computing resources; and determine a first group of computing devices and a second group of computing devices among multiple groups of computing devices based on the calculated computing utilization.
[0019] In some examples, the computational load may be the amount of computation performed within a predefined duration, the amount of available computing resources may be the maximum amount of computation that the group of computing devices can perform within the predefined duration, and the computational utilization may be the ratio of the computational load to the amount of available computing resources.
[0020] In some examples, the predefined duration may be a day, a week, a month, or a year.
[0021] In some examples, one or more processors may be configured to: determine multiple projects assigned to the group of computing devices; for each determined project: access a job log of queries for the project; and determine a project computational load for the project based on the job log of queries; and calculate a sum of the project computational loads for multiple projects.
[0022] In some examples, one or more processors may be configured to determine a transfer of one or more items between multiple groups of computing devices to reduce a computing utilization difference between a first group of computing devices and a second group of computing devices based on at least one of: the transfer reduces computing utilization of the first group of computing devices; or the transfer increases computing utilization of the second group of computing devices.
[0023] In some examples, one or more processors may be configured to store an assignment map indicating a mapping of items to groups of computing devices based on a determined transfer of one or more items between the groups of computing devices.
[0024] In some examples, the one or more processors may be configured to: determine one or more secondary cost factors associated with transferring one or more items between multiple groups of computing devices; assign a first weighted value to a computing utilization difference and assign a corresponding weighted value to each of the one or more secondary factors; and calculate a sum of the weighted computing utilization difference and the weighted secondary cost factors. The determined transfer of the one or more items between the multiple groups of computing devices may reduce the calculated sum.
[0025] In some examples, the one or more secondary cost factors may include at least one of: network bandwidth costs to transfer one or more items between groups of computing devices; or network communication costs to transmit data between groups of computing devices.
[0026] In some examples, the one or more processors may be configured to determine the transfer of one or more items between the multiple groups of computing devices by one of a naive solver, a greedy algorithm solver, or a mixed integer problem solver. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] Figure 1 is a block diagram of an example system according to aspects of the present disclosure.
[0028] Figure 2 is a block diagram of an example balancer according to aspects of the present disclosure.
[0029] Figure 3 is a flow chart of an example balancing routine according to aspects of the present disclosure.
[0030] Figure 4 and Figure 5yes Figure 3 A flowchart of an example subroutine of a routine. DETAILED DESCRIPTION
[0031] Overview
[0032] The present disclosure improves the balance of computing utilization between computing devices of a computing system. This is achieved using an assignment solver to analyze the computing cost of each of the computing devices of the computing system, wherein a solution that minimizes the computing differences between the computing devices can be considered as an optimized storage solution. Based on the analysis, the assignment solver can reassign items from a computing device with relatively high computing cost to a computing device with relatively low computing cost in order to achieve reduced or minimized computing differences.
[0033] In some examples, computational cost optimization can be the only factor used by the assignment solver to balance storage between multiple computing devices. In other examples, storage cost optimization can be one of multiple factors used to balance storage. For example, improvements to computational differences between computing devices can offset the cost of moving items from one computing device to another computing device. Additionally or alternatively, improvements to computational differences between computing devices can offset communication costs caused by assigning items of a shared data set to separate computing devices. The assignment solver can balance these factors along with other factors to achieve more comprehensive cost optimization for both storage and communication at the computing system.
[0034] While the concepts described herein may be used to manage data across multiple computing devices, it should be understood that such concepts may also be applied to other network architectures, such as those that utilize cells.
[0035] Example System
[0036] Figure 1 An example system 100 including a distributed computing environment is shown. Multiple data centers 160, 170, 180 can be communicatively coupled, for example, via a network 150. The data centers 160, 170, 180 can also communicate with one or more client devices, such as a client 110, via the network 150. Thus, for example, the client 110 can perform operations in the "cloud." In some examples, the data centers 160, 170, 180 can also communicate with a controller 190.
[0037] Data centers 160 to 180 may be located relatively far from each other. For example, data centers may be located in various countries around the world. Each data center 160, 170, 180 may include one or more computing devices, such as processors, servers, slices, units, etc. For example, Figure 1As shown, data center 160 includes computing devices 162, 164, data center 170 includes computing device 172, and data center 180 includes computing devices 181 to 186. Programs can be executed across these computing devices, for example, so that some operations are performed by one or more computing devices in the first data center, and other operations are performed by one or more computing devices in the second data center. In some examples, the computing devices in each data center can have different capacities. For example, different computing devices can have different processing speeds, workloads, etc. Although only a few of these computing devices are shown, it should be understood that each data center 160, 170, 180 can include any number of computing devices, and the number of computing devices in the first data center can be different from the number of computing devices in the second data center. In addition, it should be understood that the number of computing devices in each data center 160 to 180 can change over time, such as with the removal, replacement, upgrade, or expansion of hardware.
[0038] In some examples, each data center 160 to 180 may also include multiple storage devices (not shown), such as hard drives, random access memories, disks, disk arrays, tape drives, or any other type of storage devices. The storage devices may implement any of a variety of architectures and technologies, including but not limited to direct attached storage (DAS), network attached storage (NAS), storage area networks (SAN), Fibre Channel (FC), Fibre Channel over Ethernet (FCoE), hybrid architecture networks, etc. In addition to storage devices, the data center may also include many other devices, such as cables, routers, etc. In addition, in some examples, the data centers 160 to 180 may be virtualized environments. In addition, although only a few data centers 160 to 180 are shown, many data centers may be coupled via the network 150 and / or additional networks.
[0039] In some examples, controller 190 can communicate with computing devices in data centers 160 to 180 and can facilitate the execution of programs. For example, controller 190 can track the capacity, state, workload, or other information of each computing device and use such information to assign tasks. Similar to client 110 described above, controller 190 can include processor 198 and memory 192, including data 194 and instructions 196. In other examples, such operations can be performed by one or more of the computing devices in one of data centers 160 to 180, and a separate controller can be omitted from the system.
[0040] Each client 110 can be, for example, a computer intended for use by an individual or entity. Client 110 can have all the internal components common in a personal computer, such as a central processing unit (CPU), a CD-ROM, a hard disk drive, and a display device (e.g., a monitor with a screen, a projector, a touch screen, a small LCD screen, a television, or another device, such as an electrical device operable to display the information processed by the processor 120), a speaker, a modem and / or a network interface device, a user input such as a mouse, a keyboard, a touch screen, or a microphone, and all components for connecting these elements to each other. In addition, a computer according to the system and method described herein can include a device capable of processing instructions and transmitting data to humans and other computers (including general-purpose computers, PDAs, tablet computers, mobile phones, smart watches, network computers lacking local storage capabilities, television set-top boxes, and other networking devices) and transmitting data from humans and other computers.
[0041] The client 110 may include a processor 120, a memory 130, and other components typically found in a general-purpose computer. The memory 130 may store information accessible by the processor 120, including instructions 132 that may be executed by the processor 120. The memory may also include data 134 that may be retrieved, manipulated, or stored by the processor 120. The memory 130 may be a type of non-transitory computer-readable medium capable of storing information accessible by the processor 120, such as a hard drive, a solid-state drive, a tape drive, an optical storage device, a memory card, a ROM, a RAM, a DVD, a CD-ROM, a writable memory, and a read-only memory. The processor 120 may be a well-known processor or other less well-known types of processors. Alternatively, the processor 120 may be a dedicated controller, such as an ASIC.
[0042] The instructions 132 may be a set of instructions that are directly executed by the processor 120, such as machine code, or a set of instructions that are indirectly executed, such as a script. In this regard, the terms "instructions," "steps," and "programs" are used interchangeably herein. The instructions 132 may be stored in an object code format for direct processing by the processor 120, or in other types of computer languages, including scripts or sets of independent source code modules that are interpreted on demand or compiled in advance.
[0043] Data 134 may be retrieved, stored, or modified by processor 120 in accordance with instructions 132. For example, although the systems and methods are not limited to a particular data structure, data 134 may be stored in a computer register, in a relational database as a table with a plurality of different fields and records, or in an XML document. Data 134 may also be formatted in a computer readable format, such as, but not limited to, binary values, ASCII, or Unicode. In addition, data 134 may include information sufficient to identify the relevant information, such as a number, descriptive text, proprietary code, pointer, reference to data stored in other storage (including other network locations), or information used by a function to calculate the relevant data.
[0044] Application 136 can be used for any of a variety of operations. Application 136 can be downloaded, executable from instructions 132, or accessed remotely, for example. In some examples, the application can be executed remotely. For example, an application on a client device can be executed in the cloud.
[0045] although Figure 1 The processor 120 and memory 130 are functionally shown as being located within the same box, but the processor 120 and memory 130 may actually include multiple processors and memories that may or may not be stored within the same physical housing. For example, some of the instructions 132 and data 134 may be stored on a removable CD-ROM, while others may be stored within a read-only computer chip. Some or all of the instructions and data may be stored in a location physically remote from the processor 120 but still accessible to the processor. Similarly, the processor 120 may actually include a collection of processors that may or may not operate in parallel.
[0046] The client 110, the data center 160 to 180, and the control 190 may be capable of direct and indirect communication, such as communication through the network 150. For example, using an Internet socket, the client 110 may be connected to a service operating on a remote server through the Internet protocol suite. The server may establish a listening socket that may accept initiated connections for sending and receiving information. The network 150 and the intermediate nodes may include various configurations and protocols, including the Internet, the World Wide Web, an intranet, a virtual private network, a wide area network, a local area network, a private network using one or more company-specific communication protocols, Ethernet, WiFi (e.g., 702.71, 702.71b, g, n or other such standards), and HTTP and various combinations of the foregoing. Such communication may be facilitated by devices capable of transmitting data to and from other computers, such as modems (e.g., dial-up, cable, or optical fiber) and wireless interfaces.
[0047] Figure 2An example architecture of a system 200 for optimizing the assignment of items to computing devices within a distributed network is shown. The system 200 may reside on, for example, Figure 1 On a computing device in one of the data centers 160 to 180, Figure 1 or some combination thereof. Projects assigned to the system may include instruction sets, interfaces, settings, or combinations thereof for accessing data and utilizing computing resources of a distributed network, wherein resources may be organized according to projects. System 200 includes a load balancer 220, which may be responsible for assigning reservations to computing servers in a balanced manner. For example, load balancer 220 may balance the distribution of tasks within a data center or among a globally distributed system. For example, load balancer 220 may calculate explicit project-to-server assignment mappings to achieve appropriate load balancing. It may handle static query routing when all servers are available, and it may handle dynamic query routing when servers are exhausted. In addition, load balancer 220 may provide a robust method for publishing these assignments to all components in a data analysis engine that use this information.
[0048] The load balancer 220 may include a dispatch planner 225, which may run as a subcomponent in the load balancer 220. For example, the dispatch planner 225 may be a module, such as a C++ or other type of module. In some instances, the dispatch planner 225 may replace an existing routing calculation component. While in some examples, a single dispatch planner may generate dispatches for each region, in other examples, additional or fewer dispatch planners may be implemented. For example, a dispatch planner may generate dispatches for a globally distributed system, multiple dispatch planners may be implemented in a given region, and so on.
[0049] The assignment planner 225 may be responsible for computing and publishing project-to-server assignment mappings or assignments. When computing assignments, the assignment planner 225 considers a variety of factors, such as the hierarchy of the organization and the projects to be assigned, on-demand computing consumption, storage (e.g., by organization), server storage and computing capacity, etc. In some examples, the assignment planner 225 may also consider computing reservations, such as resources guaranteed to specific customers. For example, in order to obtain the latest reservation information, the assignment planner 225 may subscribe to periodic updates, receive push notifications when new reservations are made, etc. The information considered by the assignment planner 225 when generating a new assignment scheme may include historical query information, such as metadata snapshots, data sets, etc.
[0050] The assignment planner 225 can periodically fetch data from sources and predict future events, thereby taking such predictions into account when determining assignments. For example, the assignment planner 225 can predict a drop, a customer about to send a large amount of data, a potential reservation change, etc. Such events may affect server size, project storage, or other constraints.
[0051] The assignment planner 225 may include a number of components, such as an input collector / analyzer 232 and one or more solvers 236. It should be understood that these components are merely examples, and that additional, fewer, or different components may be included in the assignment planner 225.
[0052] The input collector / analyzer 232 collects and analyzes the input data 210 of the system. The input data 210 may include computing device data 250, from which the project assignment redistribution between computing devices can be determined. Such computing device data 250 can be provided from each of the computing devices of the system. For each computing device, the computing device data 250 may include a job log 260 of queries 262. Each query 262 included in the job log 260 may also indicate the amount of computing resource consumption 264 associated with the query. The computing resource consumption 264 may be the computing load of the query 262 logged, where the computing load is measured by the amount of computing resources consumed when executing the query. One way to measure the computing load is in terms of CPU-ms, where "CPU" refers to an atomic unit of computing resources when executing a query at a query engine, and one CPU-ms indicates that one atomic unit of computing resources is used in one millisecond. For each computing device, the computing device data 250 may also indicate the available computing resources 270 at the computing device. Available computing resources 270 may be quantified using the same or similar units as computing resource consumption 264 of query 262 executed at a computing device, so that computing resource consumption 264 and available computing resources 270 may be compared to one another.
[0053] The input data 210 may also include additional data 280 related to other factors that the redistribution of project assignments between computing devices may be based on. One example factor is the cost of transferring project assignments between computing devices, because the transfer requires data to be transferred from one computing device to another computing device. Another example factor is a project cluster. If projects are executed on a common data set, the projects can be clustered, where those common data sets can be stored at or near the computing devices of the project cluster. If reassigning projects from one computing device to another computing device causes the projects executed on the same data set to be unable to be clustered, this may be detrimental to reassignment. On the contrary, if reassigning projects from one computing device to another computing device causes the projects executed on the same data set to be clustered, this may be conducive to reassignment.
[0054] The input collector / analyzer 232 of the assignment planner 225 may receive these and other project inputs from a metadata snapshot of each project. The metadata may include additional information used to determine the arrangement of project assignments, such as organizational information for each project, reservation information for the project, both soft and hard grouping of datasets and projects, dependencies between datasets and projects based on historical data (such as query history), and the like.
[0055] The assignment planner 225 can use a solver 236 to calculate feasible assignments. The solver can be, for example, a naive solver, a greedy algorithm solver, an optimization solver (such as a mixed integer problem solver), or any other type of solver. The solver 236 accepts input and runs an assignment algorithm. The assignment solver 236 can output a set of item assignments based on the received input 210.
[0056] The assignment planner 225 may output the set of project assignments determined by the solver 236 to an assignment map 240 to map the determined location assignments of the projects executed in the distributed system 100. The assignment map 240 may be used by an assignment redistribution controller 230 to redistribute the projects across the distributed network according to the assignment map 240. The redistribution may involve moving the projects from one computing device to another computing device, creating copies of the projects, etc. The assignment redistribution controller 230 may be implemented as a library in the load balancer 220.
[0057] The assignment planner 225 can run continuously, on a schedule, on demand, or in response to environmental changes or configuration changes. It can operate in a dry-run mode, in which it generates assignments but does not apply them, or in an autonomous mode, in which it automatically sends newly generated assignments to the assignment redistribution controller 230 for immediate application.
[0058] Example Method
[0059] Combination Figures 3 to 5 The diagram of Figure 2 It should be understood that the routines described herein are merely examples, and in other examples, certain steps may be added, subtracted, replaced, or reordered.
[0060] Figure 3The example routine 300 of involves minimizing the difference in computing utilization between a computing device of a system having the highest computing utilization level and a computing device of a system having the lowest computing utilization level. This can be achieved by determining the computing utilization difference and determining an optimized storage solution for items in the system that minimizes or at least reduces the computing utilization difference by using an assignment solver. Then, the items can be transferred between the computing devices according to the optimized storage solution.
[0061] At block 310 , for a given computing device of the system, one or more processors may determine the total computing resource consumption of the computing device. Figure 4 The subroutine 400 of depicts an example implementation of such determination. In the example subroutine 400, at box 410, one or more processors access the job log of the query of each project executed by the computing device. The log identifies the amount of computing resources used by each query. At box 420, one or more processors determine the amount of computing resources of the computing device used within a predetermined time window based on the job log. In this regard, the total computing resource consumption can be the sum of the resources consumed by each of the queries in the job log of the computing device within a predetermined time period. The time period can be selected to conform to the time period pattern of computing utilization, such as a period of one day, one week, two weeks, one month, one year, etc. In many systems, the daily and weekly patterns are most obvious due to the low utilization during night and weekends, respectively. In other systems, the monthly, quarterly or annual patterns may be obvious because the organization executes projects related to end-of-month, end-of-quarter or end-of-year analysis, respectively.
[0062] Return to Figure 3 , operations may continue at block 320 where, for the same given computing device, one or more processors may determine computing resource utilization based on total resource consumption and available computing resources of the computing device. Figure 4 Subroutine 400 of depicts an example implementation of such a determination. In the example subroutine 400, at block 430, one or more processors may divide the amount of computing resources used by the computing device within a predetermined time window by the available computing resources of the computing device within the predetermined time window. The result may be a percentage of the consumption of the available resources. Since different computing devices of a system may have different amounts of available computing resources, characterizing utilization as a percentage allows for a more consistent comparison of utilization between computing devices. However, for systems in which all computing devices have the same or similar amounts of available computing resources, total resource consumption may be relied upon rather than computing resource utilization.
[0063] Blocks 310 and 320 may be performed for each of the computing devices of the system so that the load at each computing device may be determined separately. This results in the assignment planner calculating a plurality of resource utilization values. The plurality of resource utilization values may be stored in a memory for further processing. In some examples, the resource utilization values may be calculated by the assignment planner, but in other examples, these values may be calculated independently from the assignment planner, such as at each computing device, and then provided to the assignment planner for further processing.
[0064] Once the computational load of each of the computing devices is calculated, the operation may proceed to block 330, where one or more processors may determine the highest computational resource utilization among the plurality of computing devices. This may involve identifying the maximum computational resource utilization value from block 320 from the plurality of resource utilization values calculated in the storage device. Alternatively, only the maximum value may be stored without storing the resource utilization value, and the maximum value may be replaced when a larger value is calculated. At block 340, one or more processors of the assignment planner may determine the lowest computational resource utilization among the plurality of computing devices. This may involve identifying the minimum computational resource utilization value from block 320 from the plurality of resource utilization values calculated in the storage device. Alternatively, only the minimum value may be stored without storing the resource utilization value, and the minimum value may be replaced when a smaller value is calculated.
[0065] The difference between the determined highest computing resource utilization and the determined lowest computing resource utilization may indicate an overall resource utilization discrepancy within the computing system, which may itself indicate that computing resources are not properly balanced.
[0066] For example, consider a system including a first computing device, a second computing device, and a third computing device with computing resource utilizations of 92%, 32%, and 18%, respectively. The maximum computing resource utilization is 92% for the first device, while the minimum computing resource utilization is 18% for the third device, and the calculated difference is 74%.
[0067] At block 350, one or more processors may determine a redistribution of one or more items stored at the system that reduces the difference in computing resource utilization. In some instances, reducing the difference may involve determining a transfer that increases the computing resource utilization of a computing device with the lowest computing resource utilization among a plurality of computing devices. Additionally or alternatively, reducing the difference may involve determining a transfer that reduces the computing resource utilization of a computing device with the highest computing resource utilization among a plurality of computing devices. Additionally or alternatively, one or more processors may directly track the difference between the highest computing resource utilization and the lowest computing resource utilization, and the transfer may reduce the difference between the highest computing resource utilization and the lowest computing resource utilization. As described herein, the redistribution may be determined by an assignment planner, which may include a solver operating according to an algorithm, such as a greedy solver, a mixed integer solver, or other known solvers. The redistribution may involve moving at least one item from one computing device to another computing device, such as from the most underutilized computing device to the most overutilized computing device. Additionally or alternatively, the redistribution may involve adding at least one new item to at least one of the computing devices of the system, such as to the most underutilized computing device.
[0068] For example, continuing with the above example, the assignment planner may move a project from a first computing device with a utilization of 92% to a third computing device with a utilization of 18%. This may result in computing resource utilizations of 52%, 32%, and 55%, respectively, for example. Then, the difference between the maximum utilization of 55% and the minimum utilization of 32% is only 23%, which is a decrease from 74%, meaning that after the redistribution, the load at the computing devices is more appropriately balanced. It should be recognized that if more of the total computing resources were allocated to the third computing device, a 37% increase in utilization at the third computing device may correspond to a 40% decrease in utilization at the first computing device, because in this example, utilization is measured as a percentage rather than as a total computing resource consumed.
[0069] At block 360, the one or more processors may instruct the storage controller to perform the determined redistribution. The storage controller may then perform the instructed redistribution, wherein the items may be redistributed from one computing device to another computing device and the items may be transferred across a network connecting the computing devices of the system in accordance with the redistribution. The redistribution may result in a reduction in variance in computing utilization across the system.
[0070] At box 370, the redistribution of the project assignment can be logged. Logging can be performed at a routing map, which can be stored in the assignment planner or stored in a memory separate from the assignment planner. The routing map can indicate the corresponding location of each project stored in the system, and the storage location of the data set that the assigned project can operate against. Operations in the system can be performed according to the routing information stored in the routing map. Therefore, when determining to relocate a project according to the redistribution of the assignment planner, updating the log map to reflect that the redistribution can avoid communication errors in the system during the operation of one or more reallocated projects.
[0071] Figure 3 The routine 300 generally describes that the assignment planner uses computational utilization parity as the only factor to rebalance the computational load across multiple computing devices. However, in other examples, computational utilization parity can be one of the multiple factors that the assignment solver of the assignment planner optimizes the project assignment distribution. Figure 5 is a subroutine 500 of routine 300 that describes how routine 300 may consider other factors to determine redistribution of items across the system.
[0072] Boxes 510 and 520 of Figure 6 depict secondary inputs to the assignment planner. Box 510 represents a cost of transfer input indicating the cost for moving a project from one computing device to another computing device. Box 520 depicts a data ingestion rate input indicating the cost of transferring data from a data set targeted by the project operation for a given project assignment plan. These inputs should be understood as examples of possible inputs to the system, as the assignment planner may also receive other inputs reflecting other project assignment factors.
[0073] At box 530, one or more processors of the assignment planner can determine the sum of the received cost inputs for a given set of project assignments. Different factors within the sum can be weighted differently, such as according to their relative importance. For example, in one example system, transferring projects from one computing device to another computing device may be particularly expensive, where the cost of this factor may be higher than the cost of other factors. In another example system designed for projects with a high degree of aggregation, data extraction for operating projects may be given greater emphasis, in which case the weight of the data extraction rate cost may be increased. Other weightings of various factors can be set to suit each given system.
[0074] At block 540, the one or more processors of the assignment planner may determine a redistribution of one or more items stored at the system that reduces the overall cost of the system. Figure 5In some cases, reducing the total cost may or may not involve reducing the compute utilization variance. For example, in a given scenario, the cost savings from reducing other cost factors of the system may outweigh the cost of increasing the load imbalance. Nevertheless, in many scenarios, reducing the total cost will also result in reducing the compute utilization variance, and thus reducing the load imbalance.
[0075] The determined item redistribution may be output from block 540, and operations may continue to block 360, as combined with Figure 3 As described in routine 300.
[0076] The above examples generally describe the redistribution of one or more items between multiple separate computing devices (such as computers, servers, etc.). However, it should be understood that the same or similar principles can be applied to other network architectures, such as those that utilize "cells". For example, a cell is a network of tightly connected computing devices, for example, the network can collaborate to run processes, have shared storage, and have very low network latency and high throughput between the computing devices in the cell. Therefore, although the examples herein are mainly described for computing the computing resource consumption and utilization and other costs of a single computing device, it should be understood that the same or similar calculations can be performed on a group of computing devices (such as a cell). Similarly, although the examples herein are mainly described for assigning items to a single computing device, it should be understood that the same or similar assignment mappings can be generated and instructions can be transmitted to assign items and data sets to a group of multiple computing devices (such as a cell), and it should be understood that these examples are similarly applicable to assigning items to a group of multiple computing devices (such as a cell).
[0077] For example, a system may include multiple computing units, each of which includes various computing devices and is assigned multiple projects. The computing load may vary greatly from project to project, meaning that some projects may consume more computing unit resources than other projects. The method and system of the present disclosure can achieve storage balance between units and computing resource balancing and load balancing by reassigning projects from one unit to another unit according to the principles described herein. As a result, computing load and other factors can be properly and dynamically balanced throughout the system. In short, optimized dynamic load balancing can reduce overload at a separate computing device or unit. This can result in fewer errors encountered by projects stored in the system, resulting in less downtime and failures, and an overall improved user experience.
[0078] Although the technology has been described herein with reference to specific embodiments, it should be understood that these embodiments are only illustrative of the principles and applications of the present technology. Therefore, it should be understood that various modifications may be made to the illustrative embodiments, and other arrangements may be designed without departing from the spirit and scope of the present technology as defined by the appended claims.
[0079] Most of the aforementioned alternative examples are not mutually exclusive, but can be implemented in various combinations to achieve unique advantages. Since these and other variations and combinations of the features discussed above can be utilized without departing from the subject matter defined by the claims, the foregoing description of the embodiments should be made in an illustrative manner rather than in a manner that limits the subject matter defined by the claims. As an example, the foregoing operations do not have to be performed in the exact order described above. On the contrary, the various steps can be disposed in different orders, such as in the opposite order, or simultaneously. Unless otherwise specified, steps can also be omitted. In addition, the examples described herein and the provision of clauses expressed as "such as", "including", etc. should not be interpreted as limiting the subject matter of the claims to specific examples; on the contrary, these examples are intended to illustrate only one of many possible embodiments. In addition, the same figure marks in different figures may identify the same or similar elements.
Claims
1. A method for load balancing between a plurality of groups of computing devices, each group comprising one or more corresponding computing devices, It is characterized in that The method comprises: determining, by one or more processors, a first group of computing devices having a highest computing utilization level and a second group of computing devices having a lowest computing utilization level; determining, by the one or more processors, a transfer of one or more items between the plurality of groups of computing devices that reduces a difference in computing utilization between the first group of computing devices and the second group of computing devices; and The one or more processors instruct the plurality of groups of computing devices to execute the determined transfers.
2. The method according to claim 1, It is characterized in that Determining the computing utilization density includes: For each set of computing devices: determining, by the one or more processors, a computational load of the set of computing devices; determining, by the one or more processors, an amount of available computing resources of the set of computing devices; and calculating, by the one or more processors, a computing utilization of the group of computing devices based on the computing load and the amount of available computing resources; and The first group of computing devices and the second group of computing devices are determined, by the one or more processors, based on the calculated computing utilization.
3. The method according to claim 2, It is characterized in that The computational load is the amount of computation performed within a predefined duration, wherein the amount of available computing resources is the maximum amount of computation that the group of computing devices can perform within the predefined duration, and wherein the computational utilization is the ratio of the computational load to the amount of available computing resources.
4. The method according to claim 3, It is characterized in that The predefined duration is one day, one week, one month or one year.
5. The method according to claim 3, It is characterized in that Determining the computational load includes: determining, by the one or more processors, a plurality of items to assign to the set of computing devices; For each identified project: a job log of queries accessed by the one or more processors to access the project; and determining, by the one or more processors, a project computation load of the project based on the queried job log; and A sum of the project computational loads of the plurality of projects is calculated by the one or more processors.
6. The method according to claim 1, It is characterized in that Determining the transfer of one or more items between the plurality of groups of computing devices to reduce the computing utilization difference between the first group of computing devices and the second group of computing devices comprises at least one of the following: determining that the transferring reduces the computing utilization of the first group of computing devices; or It is determined that the transferring increases the computing utilization of the second group of computing devices.
7. The method according to claim 1, It is characterized in that Also included is storing, by the one or more processors, an assignment map indicating a mapping of items to the plurality of groups of computing devices based on the determined transfer of one or more items between the plurality of groups of computing devices.
8. The method according to claim 1, It is characterized in that Also includes: determining, by the one or more processors, one or more secondary cost factors associated with transferring the one or more items between the plurality of groups of computing devices; assigning, by the one or more processors, a first weighting value to the computational utilization difference and assigning a respective weighting value to each of the one or more secondary factors; and calculating, by the one or more processors, a sum of the weighted computing utilization difference and the weighted secondary cost factor, Wherein the determined transfer of the one or more items between the plurality of groups of computing devices reduces the calculated sum.
9. The method according to claim 6, It is characterized in that The one or more secondary cost factors include at least one of the following: a network bandwidth cost to transfer the one or more items between the plurality of groups of computing devices; or Network communication costs for transferring data between the plurality of groups of computing devices.
10. The method according to claim 1, It is characterized in that Determining the transfer of one or more items between the plurality of groups of computing devices is performed by one of a naive solver, a greedy algorithm solver, or a mixed integer problem solver.
11. A system for load balancing between a plurality of groups of computing devices, each group comprising one or more corresponding computing devices, It is characterized in that The system comprises: one or more memories; and One or more processors in communication with the one or more memories, the one or more processors configured to: determining a first group of computing devices having a highest computing utilization level and a second group of computing devices having a lowest computing utilization level; determining a transfer of one or more items between the plurality of groups of computing devices that reduces a difference in computing utilization between the first group of computing devices and the second group of computing devices; and The plurality of groups of computing devices are instructed to perform the determined transfers.
12. The system according to claim 11, It is characterized in that The one or more processors are configured to: For each set of computing devices: determining a computational load of the set of computing devices; determining an amount of available computing resources of the set of computing devices; as well as calculating a computing utilization of the group of computing devices based on the computing load and the amount of available computing resources; as well as The first group of computing devices and the second group of computing devices in the plurality of groups of computing devices are determined based on the calculated computing utilization.
13. The system of claim 12, It is characterized in that The computational load is the amount of computation performed within a predefined duration, wherein the amount of available computing resources is the maximum amount of computation that the group of computing devices can perform within the predefined duration, and wherein the computational utilization is the ratio of the computational load to the amount of available computing resources.
14. The system of claim 13, It is characterized in that The predefined duration is one day, one week, one month or one year.
15. The system of claim 13, It is characterized in that The one or more processors are configured to: determining a plurality of items assigned to the set of computing devices; For each identified project: a job log of queries that accessed the project; and determining a project computing load of the project based on the queried job log; as well as A sum of the project calculation loads of the plurality of projects is calculated.
16. The system of claim 1, It is characterized in that The one or more processors are configured to determine the transfer of one or more items between the plurality of groups of computing devices to reduce the computing utilization difference between the first group of computing devices and the second group of computing devices based on at least one of the following: The transferring reduces the computing utilization of the first group of computing devices; or The transferring increases the computing utilization of the second set of computing devices.
17. The system of claim 11, It is characterized in that The one or more processors are configured to store an assignment map indicating a mapping of items to the plurality of groups of computing devices based on a determined transfer of one or more items between the plurality of groups of computing devices.
18. The system of claim 11, It is characterized in that The one or more processors are configured to: determining one or more secondary cost factors associated with transferring the one or more items between the plurality of groups of computing devices; assigning a first weighting value to the computational utilization difference and assigning a respective weighting value to each of the one or more secondary factors; and Calculate the sum of the weighted compute utilization variance and the weighted secondary cost factors, Wherein the determined transfer of the one or more items between the plurality of groups of computing devices reduces the calculated sum.
19. The system of claim 16, It is characterized in that The one or more secondary cost factors include at least one of the following: a network bandwidth cost to transfer the one or more items between the plurality of groups of computing devices; or Network communication costs for transferring data between the plurality of groups of computing devices.
20. The system of claim 11, It is characterized in that The one or more processors are configured to determine the transfer of one or more items between the multiple groups of computing devices by one of a naive solver, a greedy algorithm solver, or a mixed integer problem solver.
Citation Information
Patent Citations
Compute Load Balancing In A Distributed Environment
US20230289240A1
