System and method for optimizing dispatching and scheduling decisions for a number of computations workloads

Abstract

The invention relates to a system (V) which is configured to optimize scheduling and dispatching decisions for a number of computation workloads (W), briefly called workloads, whereby each computation workload is to be executed in a data center (DC) of a plurality of data centers (DC1, DC3) each operated by one or more providers (P), whereby the system comprises : a) an input interface for receiving workload related meta information containing completion date as well as time and resource consumption and computing capabilities; b) an aggregation unit (AG) for deriving time-depending carbon footprint of each data center from aggregated weather forecast data at a data center's location and data of every energy source type available for power production during a time period of the weather forecast; c) a dispatching unit (S) for assigning / issuing each workload to a data center and for determining energy consumption of the assigned data center when executing the workload based on the received meta information; d) a scheduling unit (S) for generating the scheduling and dispatching decisions with respect to a minimi zed carbon footprint based on the time-dependent carbon footprint and on the assignment and determination according to c); and e) an execution unit (EX) for receiving each generated scheduling and dispatching decisions and for initiating execution of such a workload at the corresponding data center.

Description

[0001] Description

[0002] System and method for optimi zing dispatching and scheduling decisions for a number of computations workloads

[0003] The invention relates to a system and a method for optimi zing dispatching and scheduling decisions for a number of computation workloads , whereby each computation workload is to be executed in a data center of a plurality of data centers each operated by one or more providers .

[0004] BACKGROUND OF THE INVENTION

[0005] Data centers consume a lot of power for computation and cooling . In turn, many of these data centers feature their own or nearby local power plants to leverage renewable energies like solar or wind power for their sustainable operation . Most of the cloud providers operating one or more data centers have strategies to reach carbon- free and net zero carbon emissions goals .

[0006] When issuing workloads to the data center of a cloud provider, it therefore makes sense to issue workloads to data centers that are currently in the region where a green energy production is being generated based on the weather conditions and on the region facilities of producing the green energy .

[0007] The term workload is also often used in the cloud computing environment . It usually means a platform-independent service or independently executable program code . A workload can be also defined as any functional unit and can be a speci fic application or service . In cloud computing, workloads are usually abstracted from the hardware on which they run . In most cases , users of a workload cannot see which physical system is providing the service . Workloads often have defined interfaces through which they receive input data and / or output data . The current ef forts of the cloud providers focus on their own internal data center locations and are not utili zing the worldwide cloud provider independent distributed data centers , nor they are using other cloud providers or take into consideration the speci f icalities of the j obs .

[0008] So sustainability aspects are not suf ficiently considered . It remains a challenge to reduce the C02 emissions caused by the workloads .

[0009] SUMMARY OF THE INVENTION

[0010] It is therefore the obj ective of the present invention to provide a method for providing improved dispatching and scheduling decisions for a number of computation workloads in order to reduce 002 emission .

[0011] The above-mentioned obj ective is achieved by a method and one or more apparatus or a system according to the features of the independent claims . Preferred embodiments of the invention are described in the dependent claims . Any combination of the features of the dependent claims to each other and with the features of the independent claims is possible .

[0012] An aspect of the invention is a system which is configured to optimi ze scheduling and dispatching decisions for a number of computation workloads , briefly called workloads , whereby each computation workload is to be executed in a data center of a plurality of data centers each operated by one or more providers , whereby the system comprises : a) an input interface for receiving workload related meta information containing completion date as well as time and resource consumption and computing capabilities ; b) an aggregation unit for deriving time-depending carbon footprint of each data center from aggregated weather forecast data at a data center' s location and data of every energy source type available for power production during a time period of the weather forecast ; c) a dispatching unit for assigning and / or issuing each workload to a data center and for determining energy consumption of the assigned data center when executing the workload based on the received meta information; d) a scheduling unit for generating the scheduling and dispatching decisions with respect to a (most ) minimi zed carbon footprint based on the time-dependent carbon footprint and on the assignment and determination according to c ) ; and e ) an execution unit for receiving each generated scheduling and dispatching decisions and for initiating execution of such a workload at the corresponding data center .

[0013] Resource consumption can be express in a maximum percentage of resource usage and computing capability needs can be expressed as percentage of CPU, GPU and / or VM load .

[0014] A scheduling decision relates to a sequence of execution tasks to complete a workload by the completion time . Tasks can be scheduled . The scheduling and dispatching decisions influence the timing and pri zing / cost ( as benefit and as input for scheduling tasks ) . Workload are distributed to the data centers . The workloads come from the user . A dispatching decision relates to a data center to which the workload or even at least one task is assigned / issued .

[0015] A complicated workload comprises multiple tasks . It could be also a 1 : 1 relationship between them . Decision relates to where and when the workload is issued . There are so-called execution tasks generated by the scheduling unit . Completed execution of these scheduled tasks ensures completion of the workload . Therefore , the system allows users to issue their workload to data centers around the globe , based on the regional predicted green energy production can make a signi ficant impact on reducing the environmental impact , lowering the costs of the execution, and increasing the sustainable execution of computation workloads .

[0016] The plurality of workload related meta information from a storage unit can be received by means of an input interface , wherein the input interface is preferably an application programming interface , API .

[0017] Meta information can further comprise workload runtime related information .

[0018] Such information can be i f the workload is interruptible . That means that a workload can be paused and can continue later on . A workload can be separable . That means it could be separated in two or more parts which can be separated executed .

[0019] Aggregating data of weather forecast and / or every energy source type available for power production during a time period of the weather forecast can be provided by the provider and / or aggregated from external (web based) sources .

[0020] Such determination of energy consumption of the assigned data center while executing the workload based on the received meta information takes into account an average percentage of carbon free energy consumed at every data center' s location on an hourly basis and / or an average operational gross emissions per unit of energy from the grid .

[0021] The input data is provided or received as input for the optimi zation step . Hence , the input data is used for optimi zing the schedule and dispatching in order to reach a minimal ecological footprint . Most popular, the ecological footprint is calculated as the C02 equivalent emission . Not only a minimi zed carbon footprint shall be reached . It is also a goal , to maximi ze usage of regenerative energies ( e . g . which minimi zes usage of nuclear power as well ) .

[0022] However, the ecological footprint is not limited to C02 . Hence other considerations and other sustainability impact categories can enter the ecological footprint . The minimal ecological footprint can be equally referred to as sustainably optimi zation obj ective or sustainability aspect .

[0023] The advantage of the present invention is that , contrary to prior art , the sustainability aspects are considered in the sense of the minimal ecological footprint .

[0024] A further advantage is that the invention is not limited to C02 emissions , but instead diverse footprints and hence sustainability aspects can be flexibly considered .

[0025] A further aspect of the invention is a computer-implemented method for optimi zing scheduling and dispatching decisions for a number of computation workloads , briefly called workloads , whereby each computation workload is to be executed in a data center of a plurality of data centers each operated by one or more providers , whereby the method comprises the following steps : a) receiving workload related meta information containing completion date as well as time and resource consumption and computing capabilities ; b) deriving time-depending carbon footprint of each data center from aggregated weather forecast data at a data center' s location and data of every energy source type available for power production during a time period of the weather forecast ; c) assigning / issuing each workload to a data center and determining energy consumption of the assigned data center when executing the workload based on the received meta information; d) generating the scheduling and dispatching decisions with respect to a minimi zed carbon footprint based on the time-dependent carbon footprint and on the assignment and determination according to c ) ; and e ) initiating execution of such a workload at the corresponding data center according to the generated scheduling and dispatching decisions .

[0026] The at least one processor of the system is preferably configured to repeat steps a ) , b ) , c ) , d) and e ) until carbon footprint requirement has been met or a certain time limit has been exceeded .

[0027] Embodiments as described above for the system can be analogous applied for the method and for computer program (product ) and for the computer-readable storage medium .

[0028] This system which can be implemented by hardware , firmware and / or software or a combination of them .

[0029] The computer-readable storage medium stores instructions executable by one or more processors of a computer, wherein execution of the instructions causes the computer system to perform the method .

[0030] The computer program (product ) is executed by one or more processors of a computer and performs the method .

[0031] BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWINGS

[0032] The foregoing and other aspects of the present invention are best understood from the following detailed description when read in connection with the accompanying drawings . For the purpose of illustrating the invention, there are shown in the drawings embodiments that are presently preferred, it being understood, however, that the invention is not limited to the speci fic instrumentalities disclosed . Included in the drawings are the following figures : Figure depicts a schematic overview over potential components of the inventive system.

[0033] DETAILED DESCRIPTION OF PREFERRED EXAMPLES / EMBODIMENTS

[0034] The figure depicts a schematic overview over potential components of the inventive system.

[0035] The system V retrieves computation workloads W from multiple and various clients C, i.e., users U and / or customers with their own server farms SV, and schedules them based on information about data centers DC1, ... , DCx, cloud providers P, as well as on electricity related Data E, e.g. electricity production maps, electricity production by an energy source and operation time / maintenance of the energy source according to weather f orecast / conditions , and on meta information about the properties of the job of the workload on one or more data centers from one or more cloud providers. Current weather and energy mix data might also influence dispatching and scheduling decisions or lead to readjusting former decisions.

[0036] It enables to optimize for sustainable and cost-efficient execution of the job while still matching other constraints like completion deadlines. Workload related meta information about the data centers includes information about their computing capabilities to run workloads like types of VMs, GPUs, big data tooling support and their physical location. It can be either retrieved from the websites and in a storage recorded / stored marketing material of cloud providers or it is provided by the cloud providers (e.g., via an API) . If the cloud provider offers the data, the system will incorporate aggregated data regarding time-dependent carbon footprint and energy source type. Such data may include the average percentage of carbon free energy consumed in a particular location on an hourly basis and the average operational gross emissions per unit of energy from the grid. If the data is not available from the cloud provider, it might be available from other sources like https: / / app.electricitymaps.com / map. The system V, for example , consists of the following components :

[0037] 1 . User interface :

[0038] The actors / users U provide their workloads / j obs together with the following meta-information via a user interface , e . g . by input data on client C : a . deadline (hard or soft ) , date and time until the workload needs to be executed and completed . In case of a soft deadline , the system can schedule the workload to exceed the deadline with a configurable percentage only i f it utili zes a higher amount of green energy, b . maximum cost - The maximum cost that the user is willing to accept to run this workload, c . required resources / computing capabilities - speci fic resources needed by a data center to run the workload, like GPUs or high scaled VMs or machine learning capabilities . d . interruptible - is the workload capable to be interrupted for the system to switch the execution across various data centers to increase the execution with green energy . e . time duration - an estimated time duration of the j ob .

[0039] 2 . Power Production aggregation unit AG, also called Aggregator

[0040] This unit AG aggregates the data received from multiple inputs to compute a clear timetable of when, where how much and for how long does energy with low carbon footprint is being produced . This subsystem will save and update a system database DB periodically the timetable with the location and time-dependent carbon footprint generated for each location where a data center is registered within the system and where weather forecast data are available. Energy production at the data center's location can be determined by: a. Electricity production map data

[0041] These are applications and APIs that collect data on where (as location) and with what carbon footprint is generated. An example is: www.electricitymaps.com b. Electricity production by energy source

[0042] APIs that collect data on power production breakdown based on the types of production (hydro, geothermal, nuclear, solar, wind, gas, coal, etc...) . c. Planned opera tion / maintenance and weather forecasts Inputs received from the cloud providers or from other sources and from weather forecast's systems to allow the system to update its timetable with locations and data centers that cannot pick-up workload for a specific timeframe. Scheduler S

[0043] The scheduler comprises a dispatching unit and a scheduling unit and / or any other kind of combination of them. The Scheduler uses information provided by the aggregation unit AG, through the system database DB, and the workloads pushed e.g. by the users to generate / determine dispatching and scheduling decisions when and where the workloads should be assigned / issued (e.g., one or more data centers) . The scheduler creates and schedules necessary execution tasks to complete the workload. These tasks are saved in the system database DB and are being picked-up by an execution unit Ex, also called Executor.

[0044] The scheduling process also determines the carbon footprint produced for the execution of the workload. Energy consumption of the workloads' execution of the assigned data center (s) can be provided by the provider (s) . Such carbon footprint can also be estimated via the scheduling process by using so-called SCI (Software Carbon Intensity specification, https : / / www . thoughtworks . com / insights / blog / ethical- tech / calculating-sof tware-carbon-intensity ) based on the following equation derived from the equation of the above mentioned URL:

[0045] SCI = (E*I) per R, where:

[0046] E - energy consumption

[0047] I - Location based carbon intensity

[0048] R= per j ob / calculation (any unit) , e.g. hour [12:00] If the workload is interruptible, the scheduler will analyze possible interruptions to decrease the total carbon footprint. Alongside maximizing the green energy (in other words: minimizing the carbon footprint) used to execute the workloads, it also considers the deadlines, time duration and required resources and maybe also maximum costs. In case of new relevant information coming from the aggregation unit, the scheduler can update the execution tasks and notify an in the following explained execution unit Ex to pick-up the new schedule. Execution unit EX

[0049] The execution unit, also called executor, picks-up the scheduling and dispatching decisions e.g. in form of execution tasks generated by the scheduler. The executor initiates execution of a workload at the corresponding data center and ensures that the execution tasks are completely executed.

[0050] In case of interruptible workloads, it is responsible with pausing the execution of the workload in a data center, gathering the intermediate results and moving the execution to another data center based on the scheduling done by the scheduler. Cloud Providers P with Data Centers DC

[0051] The actual workload of the actors / users is eventually scheduled on data centers of external cloud providers. Adapters A for individual cloud providers

[0052] The system features different adapters for each of the cloud providers P . They enable assigning / issuing workloads to those cloud providers and their individual data centers . In addition, they can also enable the retrieval of information about the individual data centers like estimated or measured average percentage of carbon free energy consumed in a particular location on an hourly basis , the average operational gross emissions per unit of energy from the grid or the available resources for computations . Moreover, it may also include online / up-to- date information like the current power consumption / production measurements , weather, or computation resources that are still available .

[0053] 7 . System database DB

[0054] A system database stores all information about the cloud providers , their data centers , their compute resources and capabilities , the power production timelines of these data centers . This information was either retrieved from the cloud providers and the API s or it was added manually ( e . g . , by analyzing websites or marketing material of the cloud providers ) .

[0055] The method can be executed by at least one processor such as a microcontroller or a microprocessor, by an Application Speci fic Integrated Circuit (AS IC ) , by any kind of computer, including mobile computing devices such as tablet computers , smartphones , or laptops , or by one or more servers in a control room or cloud .

[0056] For example , a processor, controller, or integrated circuit of the system and / or computer and / or another processor may be configured to implement the acts described herein .

[0057] The above-described method may be implemented via a computer program (product ) including one or more computer-readable storage media having stored thereon instructions executable by one or more processors of a computing system . Execution of the instructions causes the computing system to perform operations corresponding with the acts of the method described above .

[0058] The instructions for implementing processes or methods described herein may be provided on non-transitory computer- readable storage media or memories , such as a cache , buf fer, RAM, FLASH, removable media, hard drive , or other computer readable storage media . A processor performs or executes the instructions to train and / or apply a trained model for controlling a system . Computer readable storage media include various types of volatile and non-volatile storage media . The functions , acts , or tasks illustrated in the figures or described herein may be executed in response to one or more sets of instructions stored in or on computer readable storage media . The functions , acts or tasks may be independent of the particular type of instruction set , storage media, processor or processing strategy and may be performed by software , hardware , integrated circuits , firmware , micro code and the like , operating alone or in combination . Likewise , processing strategies may include multiprocessing, multitasking, parallel processing and the like .

[0059] The invention has been described in detail with reference to embodiments thereof and examples . Variations and modi fications may, however, be af fected within the spirit and scope of the invention covered by the claims . The phrase " at least one of A, B and C" as an alternative expression may provide that one or more of A, B and C may be used .

[0060] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments of the invention . As used herein, the singular forms "a" , "an" , and "the" are intended to include the plural form as well , unless the context clearly indicates otherwise . It is to be understood that the elements and features recited in the appended claims may be combined in di f ferent ways to produce new claims that likewise fall within the scope of the present invention . Thus , whereas the dependent claims appended below depend from only a single independent or dependent claim, it is to be understood that these dependent claims may, alternatively, be made to depend in the alternative from any preceding or following claim, whether independent or dependent , and that such new combinations are to be understood as forming a part of the present speci fication .

[0061] None of the elements recited in the claims are intended to be a means-plus- function element unless an element is expressly recited using the phrase "means for" or, in the case of a method claim, using the phrases "operation for" or " step for" .

[0062] While the present invention has been described above by reference to various embodiments , it should be understood that many changes and modi fications may be made to the described embodiments . It is therefore intended that the foregoing description be regarded as illustrative rather than limiting, and that it be understood that all equivalents and / or combinations of embodiments are intended to be included in this description .

Claims

Patent claims1 . System (V) which is configured to optimi ze scheduling and dispatching decisions for a number of computation workloads (W) , briefly called workloads , whereby each computation workload is to be executed in a data center ( DC ) of a plurality of data centers ( DC1 , DC3 ) each operated by one or more providers ( P ) , whereby the system comprises : a) an input interface for receiving workload related meta information containing completion date as well as time and resource consumption and computing capabilities ; b) an aggregation unit (AG) for deriving time-depending carbon footprint of each data center from aggregated weather forecast data at a data center' s location and data of every energy source type available for power production during a time period of the weather forecast ; c) a dispatching unit ( S ) for assigning / issuing each workload to a data center and for determining energy consumption of the assigned data center when executing the workload based on the received meta information; d) a scheduling unit ( S ) for generating the scheduling and dispatching decisions with respect to a minimi zed carbon footprint based on the time-dependent carbon footprint and on the assignment and determination according to c ) ; and e ) an execution unit (EX ) for receiving each generated scheduling and dispatching decisions and for initiating execution of such a workload at the corresponding data center .2 . System according to claim 1 , wherein a scheduling decision relates to a sequence of execution tasks to complete a workload by the completion time and a dispatching decision relates to a data center to which the workload is assigned .3 . System according to claim 1 or 2 , wherein determination according to c ) takes into account an average percentage ofcarbon free energy consumed at every data center s location on an hourly basis and / or an average operational gross emissions per unit of energy from the grid .4 . System according to any of the previous claims , wherein determination according to c ) takes into account estimated percentage of carbon- free energy consumption from the energy source type data and weather forecast data, which can be provided by the provider and / or aggregated from external sources .

5. System according to any of the previous claims , wherein meta information further comprises workload runtime related information .

6. Computer-implemented method for optimi zing scheduling and dispatching decisions for a number of computation workloads (W) , briefly called workloads , whereby each computation workload is to be executed in a data center ( DC ) of a plurality of data centers ( DC1 , ..., DCS ) each operated by one or more providers ( P ) , whereby the method comprises the following steps : a) receiving workload related meta information containing completion date as well as time and resource consumption and computing capabilities ; b) deriving time-depending carbon footprint of each data center from aggregated weather forecast data at a data center' s location and data of every energy source type available for power production during a time period of the weather forecast ; c) assigning / issuing each workload to a data center and determining energy consumption of the assigned data center when executing the workload based on the received meta information; d) generating the scheduling and dispatching decisions with respect to a minimi zed carbon footprint based on the time-dependent carbon footprint and on the assignment and determination according to c ) ; ande ) initiating execution of such a workload at the corresponding data center according to the generated scheduling and dispatching decisions .7 . Method according to claim 6 , wherein determination according to c ) takes into account an average percentage of carbon free energy consumed at every data center' s location on an hourly basis and / or an average operational gross emissions per unit of energy from the grid .8 . Method according to any of the previous claims 6 or 7 , wherein determination according to c ) takes into account estimated percentage of carbon- free energy consumption from the energy source type data and weather forecast data, which can be provided by the provider and / or aggregated from external sources .

9. Computer program which is being executed by one or more processors of a computer in order to perform a method according to any of the previous method claims .10 . Computer readable storage media with a computer program according to the previous claim .

Images