CN109788061A - Calculating task dispositions method and device - Google Patents
Calculating task dispositions method and device Download PDFInfo
- Publication number
- CN109788061A CN109788061A CN201910064110.7A CN201910064110A CN109788061A CN 109788061 A CN109788061 A CN 109788061A CN 201910064110 A CN201910064110 A CN 201910064110A CN 109788061 A CN109788061 A CN 109788061A
- Authority
- CN
- China
- Prior art keywords
- task
- calculating task
- calculating
- processing equipment
- bandwidth
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The present invention provides a kind of calculating task dispositions method and devices, this method comprises: determining the priority level of the calculating task according to the corresponding relationship of the priority assessment index of calculating task and the priority assessment index and priority level that pre-establish;According to the priority level of the calculating task, the calculating task is deployed to the processing equipment by bandwidth required for being combined according to the including at least the calculating task of the task and the communication bandwidth to processing equipment transmission data.The utilization rate that communication bandwidth can be improved in the case where avoiding data insufficient through the above scheme, to improve computational efficiency.
Description
Technical field
The present invention relates to application specific processor technical field more particularly to a kind of calculating task dispositions methods and device.
Background technique
As Moore's Law gradually fails, the promotion of general-purpose computations framework efficiency ratio becomes more and more difficult.Dedicated computing
Framework (for example, in the form of accelerator) is one of the effective way for promoting efficiency ratio.Accelerator is usually with the shape of independent board
Formula, which is embedded into existing universal computing platform, to be assisted calculating, for example, accelerator card card insertion is promoted into existing server
The calculated performance of server.In the entire system, the operation accelerated is deployed in accelerator by universal computing platform
Calculated result is read back to universal computing platform again after the completion of waiting accelerator to calculate by operation.
Large-scale computing unit can be generally integrated on accelerator, and it is very high to calculate efficiency.But since general-purpose computations are flat
Communication bandwidth between platform and accelerator is often very limited, often will appear the limitation due to communication bandwidth in calculating process
And the problem of data insufficient supply needed for causing accelerator to calculate, this further will affect the calculated performance of whole system.
The amount for being typically different transmission data required for calculating task is different.When the big calculating task of multiple data volumes concentrates on
It can cause data serious insufficient problem when handling simultaneously together;When the small calculating task of data volume concentrates in together processing
When, although will not generally cause the problem of data insufficient supply, communication band between universal computing platform and accelerator can be caused
Width waste.
Summary of the invention
In view of this, the present invention provides a kind of calculating task dispositions method and device, to avoid data insufficient
In the case where improve communication bandwidth utilization rate, to improve computational efficiency.
To achieve the goals above, the present invention uses following scheme:
In an embodiment of the invention, calculating task dispositions method, comprising:
According to pair of the priority assessment index of calculating task and the priority assessment index and priority level that pre-establish
It should be related to, determine the priority level of the calculating task;
According to the priority level of the calculating task, according to required for the task combination for including at least the calculating task
Bandwidth and the communication bandwidth that data are transmitted to processing equipment, are deployed to the processing equipment for the calculating task.
In an embodiment of the invention, calculating task disposes device, comprising:
Priority level determination unit, for being commented according to the priority assessment index of calculating task with the priority pre-established
The corresponding relationship for estimating index and priority level determines the priority level of the calculating task;
Calculating task deployment unit, for the priority level according to the calculating task, according to including at least the calculating
Bandwidth required for the task of task combines and the communication bandwidth to processing equipment transmission data, the calculating task is deployed to
The processing equipment.
In an embodiment of the invention, computer equipment, including memory, processor and storage are on a memory and can
The computer program run on a processor, the processor realize the step of above-described embodiment the method when executing described program
Suddenly.
In an embodiment of the invention, computer readable storage medium is stored thereon with computer program, the program quilt
The step of processor realizes above-described embodiment the method when executing.
Calculating task dispositions method, calculating task deployment device, computer equipment and computer-readable storage of the invention
Medium first determines the priority level of calculating task, then according to the priority level of the calculating task, and the band according to needed for calculating task
Calculating task is deployed to processing equipment by wide and transmission data communication bandwidths, due to can be real-time during deployment task
Consider calculating task needed for bandwidth, so can Dynamical Deployment calculating task, locating so as to avoid or reduce processing equipment
Data when reason task are insufficient and bandwidth waste is difficult to the problem of balancing, can be in the feelings for avoiding data insufficient supply
The utilization rate of communication bandwidth is improved under condition.Moreover, because deployment task be using the priority level of calculating task as foundation, so
The high calculating task of priority level preferentially can be deployed to processing equipment, user experience can be improved with this.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.In the accompanying drawings:
Fig. 1 is the flow diagram of the calculating task dispositions method of one embodiment of the invention;
Fig. 2 is the flow diagram of the calculating task dispositions method of another embodiment of the present invention;
Fig. 3 is the flow diagram of the calculating task dispositions method of yet another embodiment of the invention;
Fig. 4 is to obtain calculating task according to processing equipment progress calculating task deployment simulation in one embodiment of the invention
The method flow schematic diagram of bandwidth needed for combining;
Fig. 5 is the structural schematic diagram of the computing system of one embodiment of the invention;
Fig. 6 is the Resources on Chip configuration schematic diagram of the accelerator of one embodiment of the invention;
Fig. 7 is the structural schematic diagram of the calculating task deployment device of one embodiment of the invention;
Fig. 8 is the structural schematic diagram of the calculating task deployment device of another embodiment of the present invention.
Specific embodiment
Understand in order to make the object, technical scheme and advantages of the embodiment of the invention clearer, with reference to the accompanying drawing to this hair
Bright embodiment is described in further details.Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but simultaneously
It is not as a limitation of the invention.
For data being caused seriously to supply when the big calculating task of multiple data volumes is concentrated in together while being handled at present
Deficiency is answered, and when the small calculating task of data volume concentrates in together processing, can be caused between universal computing platform and accelerator
The problem of communication bandwidth wastes, the present invention provides a kind of calculating task dispositions methods, can be deployed to and be added by dynamic adjustment
The calculating task of fast device by the control of data bandwidth needed for accelerator within the scope of communication bandwidth, and can be reduced or avoid to communicate
Bandwidth waste.
Fig. 1 is the flow diagram of the calculating task dispositions method of one embodiment of the invention.As shown in Figure 1, some implementations
The calculating task dispositions method of example, it may include:
Step S110: according to the priority assessment index of calculating task and the priority assessment index pre-established and preferentially
The corresponding relationship of grade determines the priority level of the calculating task;
Step S120: according to the priority level of the calculating task, according to the task groups for including at least the calculating task
Bandwidth required for closing and the communication bandwidth to processing equipment transmission data, are deployed to the processing for the calculating task and set
It is standby.
Above-mentioned steps S110 and step S120 can be based on existing software desk Implementation, for example, universal computing platform, it can
With using a variety of different hardware equipments realize, for example, physical server, Cloud Server, personal computer, mobile terminal equipment,
Electronic equipment etc..The calculating task can be the task to be calculated for being real-time transmitted to universal computing platform, for example, to logical operation
Sequence data, or can be to be calculated of the memory spaces such as the hard disk for being stored in advance in universal computing platform or caching
Business.Data can be transmitted between those hardware devices and above-mentioned processing equipment with certain communication bandwidth.Above-mentioned processing equipment can
To be the various equipment that can be handled the data come from the transmission of those hardware devices, for example, various application specific processors
Chip etc..The chip of the application specific processor can be realized in the form of accelerator, can be used for carrying out the place of various logic operation
Reason.
In above-mentioned steps S110, which can be the various calculating task urgency levels that can distinguish
Information, for example, the user class etc. of the type of calculating task, calculating task.The type of calculating task can be calculating type.Tool
Body, for example, the type of the calculating task may include that web page browsing type, instant messaging type etc. in the case should be i.e.
When communication type calculating task priority level can be higher than the web page browsing type calculating task priority level.Example again
Such as, the user class of calculating task may include paying customer, non-paid user etc., and in the case, the paying customer is corresponding
Calculating task priority level can be higher than the corresponding calculating task of the non-paid user priority level.
The corresponding relationship of the above-mentioned priority assessment index and priority level pre-established can be real in the form of tables of data
It is existing, it may include multipair mutual corresponding priority assessment index and priority level.Priority assessment index and priority level can
To be one-to-one mapping relations or many-to-one mapping relations.Determining that described calculate is appointed according to the corresponding relationship
When the priority level of business, specifically, the priority assessment index of the calculating task can be extracted first, then according to the meter
The priority assessment index of calculation task Query priority evaluation index in the tables of data of the corresponding relationship, then basis inquires
Priority assessment index find corresponding priority level.
In above-mentioned steps S120, the priority level of the calculating task may include the first order, the second level, the third level
Deng, wherein the first order is the superlative degree, and the second level, third level etc. can be secondary advanced.Described calculate that include at least is appointed
The task combination of business can only include the calculating task of high precedence, or can simultaneously include the institute of high precedence
State the calculating task of calculating task and time high precedence.Since processing equipment (application specific processor) can have certain storage
Resource, during application specific processor handles calculating task, the data of reuse can be kept on a processing device, so
These data reused can not need to repeat to transmit from universal computing platform to application specific processor, but in processing equipment
On storage resource it is not enough in the case where, the data that part is reused may be still needed to from universal computing platform to dedicated processes
Device transmission, be based on this reason, the task of multiple calculating tasks is combined be deployed to application specific processor in the case where, task groups
Communication bandwidth needed for closing may be less than or equal to total communication when multiple calculating task is deployed to application specific processor respectively
Bandwidth.So processing equipment various calculating tasks to be dealt with and various groups between them can be obtained respectively in advance
Bandwidth needed for closing.To processing equipment transmission data communication bandwidth, for example, from universal computing platform to application specific processor (on
State processing equipment) bandwidth of transmission data, certain communication bandwidth, which can be equal to from application specific processor to universal computing platform, to be passed
The bandwidth of transmission of data.The calculating task, which is deployed to the processing equipment, to be referred to, by the distribution of computation tasks or peace
Arranging, which can also refer to the processing equipment, is transmitted to processing equipment for all distribution or the calculating task arranged properly.
In the present embodiment, the priority level of calculating task is first determined, then according to the priority level of the calculating task, and according to
Calculating task is deployed to processing equipment by the communication bandwidth of bandwidth needed for calculating task and transmission data, due in deployment task
Can consider bandwidth needed for calculating task in real time in the process, thus can Dynamical Deployment calculating task, so as to avoid or subtract
Data of few processing equipment in the task of processing are insufficient and bandwidth waste is difficult to the problem of balancing, and can avoid counting
The utilization rate of communication bandwidth is improved in the case where according to insufficient supply.Moreover, because deployment task is the priority scheduling with calculating task
Grade is foundation, so processing equipment preferentially can be deployed to priority level high calculating task, can be improved user with this
Experience.
Fig. 2 is the flow diagram of the calculating task dispositions method of another embodiment of the present invention.As shown in Fig. 2, other
In embodiment, calculating task dispositions method shown in FIG. 1, before above-mentioned steps S110, that is, according to the priority of calculating task
The corresponding relationship of evaluation index and the priority assessment index and priority level that pre-establish, determines the preferential of the calculating task
Before grade, it may also include that
Step S130: using task type as priority assessment index, all according to corresponding to the processing equipment
Service type establishes the corresponding relationship of priority assessment index and priority level.
In above-mentioned steps S130, task type may include the user class etc. of the type of calculating task, calculating task.
Universal computing platform receives or the task to be calculated of storage can include corresponding task type, institute in the form of label, field etc.
The information of task type is obtained can extract or parse from task to be calculated.It, can for a kind of processing equipment
The calculating task of fixed type is handled, the computing resource quantity that each type of task can occupy is also possible to fixed.It should
The each type of task that processing equipment can be handled can correspond to a priority level, corresponding to different types of task
Priority level can be identical or different, to can form a pair corresponding to processing equipment between task type and priority level
It is one or more to one relationship.Above-mentioned corresponding relationship can be realized in the form of tables of data.According to corresponding to the processing equipment
It, can relatively urgent journey between each task type according to corresponding to processing equipment when all task types establish corresponding relationship
Spend the value to determine corresponding priority level.The value of the priority level of different processing can be distinguish.
In the present embodiment, since the degree that the processing speed of the calculating task of different task type influences user experience can be with
It is significantly different, so using the task type of calculating task as priority assessment index, it can more targetedly, significantly
Improve the experience of user corresponding to the demanding calculating task of processing speed.
It in other embodiments, can be using the user gradation corresponding to calculating task as priority assessment index, herein
In the case of, all user gradations according to corresponding to the processing equipment establish priority assessment index and priority level
Corresponding relationship.The user gradation may include paying customer, non-paid user etc..In this way, can be special group user
Better user experience is provided.
In some embodiments, above-mentioned steps S120, that is, according to the priority level of the calculating task, according at least wrapping
The including the calculating task of the task combines required bandwidth and the communication bandwidth to processing equipment transmission data, by the calculating
Task deployment to the processing equipment method specific embodiment, it may include:
The calculating task priority level be five-star situation under, will include at least the calculating task task
Combination priority is deployed to the processing equipment, wherein bandwidth required for the task combines is less than or equal to the communication band
It is wide.
No matter in the present embodiment, priority level is divided in which way, which can refer to above-mentioned processing equipment institute energy
The calculating task of highest priority in the various calculating tasks of processing.The case where the calculating task takes communication bandwidth substantially
Under, the task combination including at least the calculating task can only include the calculating task;Or by the meter
After calculation task distributes to processing equipment, communication bandwidth, can be by other calculating tasks and the calculating there are also in remaining situation
Task distributes to processing equipment together, can make full use of communication bandwidth with this.By being five-star described by priority level
Calculating task prioritized deployment enables to the processing task preferentially to be handled to the processing equipment.
In some embodiments, it is the five-star calculating task that the task combination, which includes all priority levels, or
Person includes that at least one priority level is secondary advanced calculating task simultaneously;Wherein, described secondary advanced lower than the superlative degree.
In the present embodiment, this advanced next priority level that preferably can be the superlative degree and be located next to, or do not having
When having the calculating task for the next priority level being located next to, this advanced next priority scheduling that can be the superlative degree and be located next to
Next one priority level of grade, and so on.Priority level is five-star calculating task in addition to comprising the calculating task,
It also may include one or more other calculating tasks, in total required band that all priority levels are five-star calculating task
Width be equal to or substantially close to communication bandwidth in the case where, by being other five-star calculating tasks and the meter by priority level
For prioritized deployment to processing equipment, enable to all priority levels is that five-star calculating task is preferentially located to calculation task together
Reason.By in the case where total required bandwidth that all priority levels are five-star calculating task is less than communication bandwidth, incite somebody to action
All priority levels be five-star distribution of computation tasks to processing equipment after, communication bandwidth can also have some residues, in this feelings
Under condition, according to bandwidth situation needed for secondary advanced calculating task, time height that bandwidth and remaining communication bandwidth can be matched
Calculating task and the priority level of grade are to be deployed to processing equipment together with those five-star calculating tasks, can be sufficiently sharp with this
With communication bandwidth, communication bandwidth waste is reduced or avoided.
Fig. 3 is the flow diagram of the calculating task dispositions method of yet another embodiment of the invention.As shown in figure 3, some
In embodiment, calculating task dispositions method shown in FIG. 1, before above-mentioned steps S120, that is, according to the excellent of the calculating task
First grade, bandwidth required for being combined according to the including at least the calculating task of the task and leading to processing equipment transmission data
Letter bandwidth may also include that before the calculating task is deployed to the processing equipment
Step S140: calculating task deployment simulation is carried out according to the processing equipment, obtains appointing including at least described calculate
Bandwidth required for the task of business combines.
In above-mentioned steps S140, calculating task deployment simulation can be simulation and appoint a calculating task or multiple calculate
The case where task combination of business is deployed to processing equipment.When carrying out deployment simulation, the processing equipment can be modeled,
And different calculating tasks is individually deployed to the model that modeling obtains, or the calculating task of various various combinations is deployed to
Obtained model is modeled, and calculates the data between general-purpose platform and application specific processor in statistics under different deployment scenarios and transmits feelings
Condition, thus bandwidth needed for obtaining corresponding calculating task or task combination.The process of deployment simulation can be by voluntarily compiling
Cheng Shixian.
Since data from general computing platform needed for calculating task is usually transmitted to application specific processor (processing equipment)
In the process, it is difficult to obtain the relevant information of bandwidth needed for calculating task.It is simulated so being disposed in the present embodiment by calculating task
Bandwidth needed for the data for the calculating task for being transmitted to application specific processor from universal computing platform being assessed in advance, to solve
The problem of being difficult to bandwidth needed for obtaining calculating task at present.
Fig. 4 is to obtain calculating task according to processing equipment progress calculating task deployment simulation in one embodiment of the invention
The method flow schematic diagram of bandwidth needed for combining.As shown in figure 4, above-mentioned steps S140, that is, counted according to the processing equipment
Task deployment simulation is calculated, bandwidth required for combining including at least the task of the calculating task is obtained, it may include:
Step S141: it is modeled according to the computing resource of the processing equipment and storage resource, obtains device model;
Step S142: the task combine analog for including at least the calculating task is deployed to the device model, and is united
The amount for counting the data transmitted in the unit time to the device model obtains bandwidth required for the task combines.
In above-mentioned steps S141, the case where which can simulate the computing resource of the processing equipment and described
The case where storage resource of processing equipment.Specifically, the case where computing resource, may include the quantity of computing unit, calculate list
The parameters such as the processing capacity of member.The case where storage resource may include the number of plies of storage medium, the size of each layer storage medium,
The storage precedence case etc. of each layer storage medium.
It, can be by resulting calculating when task combination only includes a calculating task in above-mentioned steps S142
The bandwidth of required by task is as bandwidth needed for the calculating task similar with the calculating task.When task combination is comprising multiple
When calculating task, can will it is resulting needed for bandwidth as combine similar (for example, task type combine identical) with the task
Bandwidth needed for task combination.In the identical situation of type of multiple calculating task, gained task combination institute can use
Need bandwidth divided by task combine in calculate the number of task, and using calculated result as bandwidth needed for a calculating task;Or
Person, can be arranged various task combinations by way of enumerating, and the combination of various tasks, various calculating tasks are simulated deployment respectively
To processing equipment, to obtain corresponding bandwidth, and record.When needing in practical applications according to a calculating task or
When bandwidth needed for the task of multiple calculating tasks combines carries out calculating task deployment, the bandwidth and task of inquiry record can be passed through
Combination/calculating task corresponding relationship, bandwidth needed for obtaining the calculating task or task combination in practical application, to carry out
Calculating task deployment.
In the present embodiment, by being modeled to obtain according to the computing resource and storage resource of the processing equipment of actual use
Device model, and simulation is deployed to the device model respectively by various calculating tasks or task combination, it can be real by statistics
Now the obtaining a calculating task or multiple calculating tasks of the task combines required bandwidth.
In some embodiments, by taking the data transmission between universal computing platform and accelerator as an example, calculating task deployment
Method, it may include:
(A) bandwidth demand of all kinds of calculating tasks, data bandwidth needed for obtaining all kinds of calculating tasks are assessed;
(B) data bandwidth according to needed for obtaining all kinds of calculating tasks carries out deployment scheduling to the calculating task of accelerator.
In order to accurately portray data bandwidth needed for all kinds of calculating tasks run on accelerator, in above-mentioned steps
(A) in, calculating task bandwidth demand appraisal procedure, it may include following two step:
(A1) accelerator models: being modeled for computing resource on accelerator and storage resource, obtains accelerator mockup.
(A2) deployment task completes bandwidth demand assessment: different calculating tasks being deployed to step (A1) respectively and is established
In accelerator mockup, and data bandwidth needed for counting each calculating task.
For accelerator mockup described in above-mentioned steps (A1), it is primarily useful for that the rule of the computing unit of accelerator are depicted
The calculating type that mould and accelerator can be supported.When the storage architecture of accelerator is described, several layers of storages can be described and be situated between
Matter and the amount of capacity of each layer storage.
For each calculating task described in above-mentioned steps (A2), the calculating type that usual each accelerator can be supported is
Fixed, it can be by way of being enumerated according to the quantity of the computing unit of accelerator, type and connection relationship come to each
Data bandwidth needed for calculating task is analyzed.Can computation model and storage model according to described in step (A1), statistics
Data-reusing rate assessment storage resource of the task of each calculating task or multiple calculating tasks combination at accelerator end utilizes out
Situation, real data bandwidth needed for entire calculating task or task combination can be further calculated out.
In order to efficiently using the communication bandwidth between universal computing platform and accelerator, avoid accelerator data supply not
Under the premise of foot, to complete more calculating tasks, in above-mentioned steps (B), accelerator task deployment scheduling strategy, specific implementation
Mode may include steps of:
(B1) calculating task priority level is determined: according to the priority level (emergency) of the calculating task of user to calculating
Task progress grade classification, more urgent task, priority are higher.
(B2) the task combination for determining calculating task, disposes task: being obtained according to priority and assessment each
Data bandwidth needed for calculating task is disposed.
For the determination that task in step (B2) combines, the high calculating of priority level can be disposed first in task combination and is appointed
Business calculates it and occupies the bandwidth situation between general-purpose platform and accelerator, and according to remaining general-purpose platform and can accelerate
The case where device bandwidth, further disposes the lower calculating task of priority.
In the present embodiment, through the above, bandwidth needed for each calculating task can accurately be portrayed, and can
Task is efficiently dispatched and combined whether with for calculating task promptly, both guarantees high Service Quality so as to reach
Amount obtains high calculated performance again.Aiming at the problem that accelerator data insufficient supply and bandwidth waste, it can dynamically adjust and add
Fast device computational load combination makes the control of data bandwidth required for accelerator communication bandwidth range between general-purpose platform and accelerator
Within, to reach high computational efficiency.
To make those skilled in the art be best understood from the present invention, it will illustrate implementation of the invention with specific embodiment below
Mode and effect.
Fig. 5 is the structural schematic diagram of the computing system of one embodiment of the invention.Fig. 6 is the accelerator of one embodiment of the invention
Resources on Chip configuration schematic diagram.As shown in figure 5, may include universal computing platform and accelerator in the framework of computing system.
As shown in fig. 6, multiple computing units 301 and a storage unit 302 can be configured on accelerator.Institute referring to figure 5 and figure 6
Show, in the system architecture for including universal computing platform and accelerator, the bandwidth between universal computing platform and accelerator is
1024MB/s, the storage unit (as shown in Figure 2) in accelerator comprising 20 computing units and 10MB.6 calculating tasks point
It Wei not A, B, C, D, E, F.Under above-mentioned configuration data bandwidth needed for each calculating task be respectively 256MB/s, 256MB/s,
768MB/s,640MB/s,128MB/s,128MB/s;The priority level of each calculating task is respectively 1,2,2,1,2,2, number
Smaller, the priority that calculating task has is higher;Computing unit number needed for each calculating task is all 5 (to examine in this emphasis
Consider the influence of bandwidth);It is identical that each calculating task is mapped to accelerator evaluation time of falling into a trap, and is time T.
Assuming that above-mentioned 6 tasks are successively requested by the sequence of A, B, C, D, E, F, in the prior art, usually press and appoint
The request sequence of business is mapped.Calculating task A, B is mapped in accelerator first and carries out operation, due to calculating task A and B
Occupy the communication bandwidth of 512MB/s, it is available that data bandwidth only remains 512MB/s, so task C is obstructed, subsequent task all without
Method is by operation.Etc. after tasks A and B complete, calculating task C can just be mapped in accelerator and be calculated, while D and later
Task be also blocked, there is the data bandwidth of 256MB/s to be wasted at this time.Etc. tasks C complete after, task D, E, F are mapped
It is calculated into accelerator, there is 128MB/s data bandwidth to be wasted.All task computations are completed to need the time of 3T, and
And there is massive band width to be wasted, although while task D priority with higher, could be calculated in the 3T period, take
Quality of being engaged in is excessively poor.
Using accelerator scheduling/calculating task dispositions method of the embodiment of the present invention, higher calculate of priority first is appointed
Business A and D is mapped on accelerator and is calculated, and the occupied 896MB/s of communication bandwidth, also surplus 128MB/s bandwidth is available, so
Task E can be mapped in accelerator simultaneously and be calculated.Etc. above threes calculating task complete after, calculating task B, C,
Bandwidth needed for F is exactly 1024MB/s, can be calculated simultaneously.All meters can be completed within the 2T time in the above described manner
Calculation task, meanwhile, high calculating task A, D of priority can be completed within the 1T time, have unusual service quality, and fill
Divide and data transfer bandwidth is utilized.
It can be seen that method through the embodiment of the present invention, can make full use of between universal computing platform and accelerator
Data bandwidth, and service quality can be guaranteed according to calculating task priority, the performance of acceleration can be greatly improved.
Based on inventive concept identical with calculating task dispositions method shown in FIG. 1, the embodiment of the present application also provides one
Kind calculating task disposes device, as described in following example.The principle and meter solved the problems, such as due to calculating task deployment device
Calculation task deployment method is similar, therefore the implementation of calculating task deployment device may refer to the reality of calculating task dispositions method
It applies, overlaps will not be repeated.
Fig. 7 is the structural schematic diagram of the calculating task deployment device of one embodiment of the invention.As shown in fig. 7, some implementations
The calculating task of example disposes device, it may include: priority level determination unit 210 and calculating task deployment unit 220, the two is mutual
Connection.
Priority level determination unit 210, for according to the priority assessment index of calculating task and pre-establishing preferential
The corresponding relationship of grade evaluation index and priority level, determines the priority level of the calculating task;
Calculating task deployment unit 220, for the priority level according to the calculating task, according to including at least the meter
Bandwidth required for the task of calculation task combines and the communication bandwidth to processing equipment transmission data, the calculating task is disposed
To the processing equipment.
Fig. 8 is the structural schematic diagram of the calculating task deployment device of another embodiment of the present invention.As shown in figure 8, other
The calculating task of embodiment disposes device, may also include that task bandwidth determination unit 240, connects with calculating task deployment unit 220
It connects.
Task bandwidth determination unit 240 obtains at least for carrying out calculating task deployment simulation according to the processing equipment
Bandwidth required for task including the calculating task combines.
In some embodiments, the calculating task of each embodiment disposes device can include: priority corresponding relations are established single
Member is connect with above-mentioned priority level determination unit 210.
Priority corresponding relations establish unit, are used for using task type as priority assessment index, according to the processing
All task types corresponding to equipment establish the corresponding relationship of priority assessment index and priority level.
In some embodiments, above-mentioned calculating task deployment unit 220 can include: calculating task deployment module.
Calculating task deployment module, for the priority level in the calculating task be five-star situation under, will at least
Task including the calculating task combines prioritized deployment to the processing equipment, wherein band required for the task combines
Width is less than or equal to the communication bandwidth.
In some embodiments, it is the five-star calculating task that the task combination, which includes all priority levels, or
Person includes that at least one priority level is secondary advanced calculating task simultaneously;Wherein, described secondary advanced lower than the superlative degree.
In some embodiments, task bandwidth determination unit 240 can include: equipment modeling module and bandwidth determination module,
The two is connected with each other.
Equipment modeling module is set for being modeled according to the computing resource and storage resource of the processing equipment
Standby model;
Bandwidth determination module, for the task combine analog for including at least the calculating task to be deployed to the equipment mould
Type, and the amount for the data transmitted in the statistical unit time to the device model obtain bandwidth required for the task combines.
The embodiment of the present invention also provides a kind of computer equipment, including memory, processor and storage are on a memory simultaneously
The computer program that can be run on a processor, the processor realize the various embodiments described above the method when executing described program
The step of.
The embodiment of the present invention also provides a kind of computer readable storage medium, is stored thereon with computer program, the program
The step of the various embodiments described above the method is realized when being executed by processor.
In conclusion the calculating task dispositions method of the embodiment of the present invention, calculating task deployment device, computer equipment and
Computer readable storage medium first determines the priority level of calculating task, then according to the priority level of the calculating task, and according to
Calculating task is deployed to processing equipment by the communication bandwidth of bandwidth needed for calculating task and transmission data, due in deployment task
Can consider bandwidth needed for calculating task in real time in the process, thus can Dynamical Deployment calculating task, so as to avoid or subtract
Data of few processing equipment in the task of processing are insufficient and bandwidth waste is difficult to the problem of balancing, and can avoid counting
The utilization rate of communication bandwidth is improved in the case where according to insufficient supply.Moreover, because deployment task is the priority scheduling with calculating task
Grade is foundation, so processing equipment preferentially can be deployed to priority level high calculating task, can be improved user with this
Experience.
In the description of this specification, reference term " one embodiment ", " specific embodiment ", " some implementations
Example ", " such as ", the description of " example ", " specific example " or " some examples " etc. mean it is described in conjunction with this embodiment or example
Particular features, structures, materials, or characteristics are included at least one embodiment or example of the invention.In the present specification,
Schematic expression of the above terms may not refer to the same embodiment or example.Moreover, the specific features of description, knot
Structure, material or feature can be combined in any suitable manner in any one or more of the embodiments or examples.Each embodiment
Involved in the step of sequence be used to schematically illustrate implementation of the invention, sequence of steps therein is not construed as limiting, can be as needed
It appropriately adjusts.
It should be understood by those skilled in the art that, the embodiment of the present invention can provide as method, system or computer program
Product.Therefore, complete hardware embodiment, complete software embodiment or reality combining software and hardware aspects can be used in the present invention
Apply the form of example.Moreover, it wherein includes the computer of computer usable program code that the present invention, which can be used in one or more,
The computer program implemented in usable storage medium (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.) produces
The form of product.
The present invention be referring to according to the method for the embodiment of the present invention, the process of equipment (system) and computer program product
Figure and/or block diagram describe.It should be understood that every one stream in flowchart and/or the block diagram can be realized by computer program instructions
The combination of process and/or box in journey and/or box and flowchart and/or the block diagram.It can provide these computer programs
Instruct the processor of general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices to produce
A raw machine, so that being generated by the instruction that computer or the processor of other programmable data processing devices execute for real
The device for the function of being specified in present one or more flows of the flowchart and/or one or more blocks of the block diagram.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates,
Enable the manufacture of device, the command device realize in one box of one or more flows of the flowchart and/or block diagram or
The function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that counting
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer or
The instruction executed on other programmable devices is provided for realizing in one or more flows of the flowchart and/or block diagram one
The step of function of being specified in a box or multiple boxes.
Particular embodiments described above has carried out further in detail the purpose of the present invention, technical scheme and beneficial effects
Describe in detail it is bright, it should be understood that the above is only a specific embodiment of the present invention, the guarantor being not intended to limit the present invention
Range is protected, all within the spirits and principles of the present invention, any modification, equivalent substitution, improvement and etc. done should be included in this
Within the protection scope of invention.
Claims (10)
1. a kind of calculating task dispositions method characterized by comprising
It is closed according to the priority assessment index of calculating task and the priority assessment index pre-established are corresponding with priority level
System, determines the priority level of the calculating task;
According to the priority level of the calculating task, bandwidth required for being combined according to the including at least the calculating task of the task
Communication bandwidth with to processing equipment transmission data, is deployed to the processing equipment for the calculating task.
2. calculating task dispositions method as described in claim 1, which is characterized in that referred to according to the priority assessment of calculating task
The corresponding relationship of the priority assessment index and priority level marking and pre-establish, determine the calculating task priority level it
Before, further includes:
Using task type as priority assessment index, all task types according to corresponding to the processing equipment are established preferential
The corresponding relationship of grade evaluation index and priority level.
3. calculating task dispositions method as described in claim 1, which is characterized in that the priority scheduling according to the calculating task
Grade, bandwidth required for being combined according to the including at least the calculating task of the task and the communication band to processing equipment transmission data
The calculating task is deployed to the processing equipment by width, comprising:
The calculating task priority level be five-star situation under, will include at least the calculating task task combine
Prioritized deployment is to the processing equipment, wherein bandwidth required for the task combines is less than or equal to the communication bandwidth.
4. calculating task dispositions method as claimed in claim 3, which is characterized in that the task combination includes all priority schedulings
Grade is the five-star calculating task, or simultaneously include at least one priority level is time advanced calculating task;Wherein,
Described time advanced lower than the superlative degree.
5. calculating task dispositions method as described in claim 1, which is characterized in that the priority scheduling according to the calculating task
Grade, bandwidth required for being combined according to the including at least the calculating task of the task and the communication band to processing equipment transmission data
The calculating task is deployed to before the processing equipment by width, further includes:
Calculating task deployment simulation is carried out according to the processing equipment, obtains the task combination institute including at least the calculating task
The bandwidth needed.
6. calculating task dispositions method as claimed in claim 5, which is characterized in that calculate according to the processing equipment and appoint
Business deployment simulation obtains bandwidth required for combining including at least the task of the calculating task, comprising:
It is modeled according to the computing resource of the processing equipment and storage resource, obtains device model;
The task combine analog that include at least the calculating task will be deployed to the device model, and in the statistical unit time to
The amount of the data of the device model transmission, obtains bandwidth required for the task combines.
7. a kind of calculating task disposes device characterized by comprising
Priority level determination unit, for being referred to according to the priority assessment index of calculating task with the priority assessment pre-established
The corresponding relationship of mark and priority level, determines the priority level of the calculating task;
Calculating task deployment unit, for the priority level according to the calculating task, according to including at least the calculating task
Task combine required for bandwidth and the communication bandwidth to processing equipment transmission data, the calculating task is deployed to described
Processing equipment.
8. calculating task as claimed in claim 7 disposes device, which is characterized in that further include:
Task bandwidth determination unit obtains including at least institute for carrying out calculating task deployment simulation according to the processing equipment
The stating calculating task of the task combines required bandwidth.
9. a kind of computer equipment including memory, processor and stores the meter that can be run on a memory and on a processor
Calculation machine program, which is characterized in that the processor realizes any one of claim 1 to 6 the method when executing described program
Step.
10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the program is by processor
The step of any one of claim 1 to 6 the method is realized when execution.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910064110.7A CN109788061B (en) | 2019-01-23 | 2019-01-23 | Computing task deployment method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910064110.7A CN109788061B (en) | 2019-01-23 | 2019-01-23 | Computing task deployment method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109788061A true CN109788061A (en) | 2019-05-21 |
CN109788061B CN109788061B (en) | 2021-02-26 |
Family
ID=66501164
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910064110.7A Active CN109788061B (en) | 2019-01-23 | 2019-01-23 | Computing task deployment method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109788061B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112764924A (en) * | 2021-01-14 | 2021-05-07 | 城云科技(中国)有限公司 | Task scheduling method and device and electronic equipment |
WO2024093280A1 (en) * | 2022-11-01 | 2024-05-10 | 华为技术有限公司 | Task management method, apparatus and system, and communication device and storage medium |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101227416A (en) * | 2008-01-15 | 2008-07-23 | 中兴通讯股份有限公司 | Method for distributing link bandwidth in communication network |
CN101374020A (en) * | 2007-08-20 | 2009-02-25 | 中兴通讯股份有限公司 | Centralized bandwidth distribution method for relay network |
US7843943B2 (en) * | 2007-11-21 | 2010-11-30 | Nokia Corporation | Differentiation for bandwidth request contention |
CN102932935A (en) * | 2011-08-12 | 2013-02-13 | 中国移动通信集团公司 | Data transmission method and mobile terminal |
CN103701722A (en) * | 2014-01-02 | 2014-04-02 | 网神信息技术(北京)股份有限公司 | Network bandwidth allocation method, device and system |
CN105721350A (en) * | 2016-03-04 | 2016-06-29 | 深圳市共进电子股份有限公司 | Intelligent bandwidth allocation method and apparatus |
US20180024739A1 (en) * | 2016-07-22 | 2018-01-25 | Intel Corporation | Memory Sharing for Physical Accelerator Resources in a Data Center |
US20180097740A1 (en) * | 2016-10-03 | 2018-04-05 | Cisco Technology, Inc. | Estimating time duration of bandwidth availability |
CN107949020A (en) * | 2017-11-30 | 2018-04-20 | 中国联合网络通信集团有限公司 | A kind of method and system of network game business data transmission |
-
2019
- 2019-01-23 CN CN201910064110.7A patent/CN109788061B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101374020A (en) * | 2007-08-20 | 2009-02-25 | 中兴通讯股份有限公司 | Centralized bandwidth distribution method for relay network |
US7843943B2 (en) * | 2007-11-21 | 2010-11-30 | Nokia Corporation | Differentiation for bandwidth request contention |
CN101227416A (en) * | 2008-01-15 | 2008-07-23 | 中兴通讯股份有限公司 | Method for distributing link bandwidth in communication network |
CN102932935A (en) * | 2011-08-12 | 2013-02-13 | 中国移动通信集团公司 | Data transmission method and mobile terminal |
CN103701722A (en) * | 2014-01-02 | 2014-04-02 | 网神信息技术(北京)股份有限公司 | Network bandwidth allocation method, device and system |
CN105721350A (en) * | 2016-03-04 | 2016-06-29 | 深圳市共进电子股份有限公司 | Intelligent bandwidth allocation method and apparatus |
US20180024739A1 (en) * | 2016-07-22 | 2018-01-25 | Intel Corporation | Memory Sharing for Physical Accelerator Resources in a Data Center |
US20180097740A1 (en) * | 2016-10-03 | 2018-04-05 | Cisco Technology, Inc. | Estimating time duration of bandwidth availability |
CN107949020A (en) * | 2017-11-30 | 2018-04-20 | 中国联合网络通信集团有限公司 | A kind of method and system of network game business data transmission |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112764924A (en) * | 2021-01-14 | 2021-05-07 | 城云科技(中国)有限公司 | Task scheduling method and device and electronic equipment |
WO2024093280A1 (en) * | 2022-11-01 | 2024-05-10 | 华为技术有限公司 | Task management method, apparatus and system, and communication device and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN109788061B (en) | 2021-02-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110851529B (en) | Calculation power scheduling method and related equipment | |
CN109669774B (en) | Hardware resource quantification method, hardware resource arrangement method, hardware resource quantification device and hardware resource arrangement device and network equipment | |
CN104636197B (en) | A kind of evaluation method of data center's virtual machine (vm) migration scheduling strategy | |
CN110163474A (en) | A kind of method and apparatus of task distribution | |
CN104243617B (en) | Towards the method for scheduling task and system of mixed load in a kind of isomeric group | |
CN103401939A (en) | Load balancing method adopting mixing scheduling strategy | |
CN109144710A (en) | Resource regulating method, device and computer readable storage medium | |
CN102567077B (en) | Virtualized resource distribution method based on game theory | |
CN106897206A (en) | A kind of service test method and device | |
CN103309431B (en) | Dynamic frequency scalable | |
CN109842563A (en) | Content delivery network flow dispatching method, device and computer readable storage medium | |
CN104053179B (en) | A kind of C RAN system integration project platforms | |
CN110502323B (en) | Real-time scheduling method for cloud computing tasks | |
CN104536832A (en) | Virtual machine deployment method | |
CN105373432A (en) | Cloud computing resource scheduling method based on virtual resource state prediction | |
CN115134371A (en) | Scheduling method, system, equipment and medium containing edge network computing resources | |
CN109726004A (en) | A kind of data processing method and device | |
CN109788061A (en) | Calculating task dispositions method and device | |
CN103677960A (en) | Game resetting method for virtual machines capable of controlling energy consumption | |
CN109471725A (en) | Resource allocation methods, device and server | |
CN104793993A (en) | Cloud computing task scheduling method of artificial bee colony particle swarm algorithm based on Levy flight | |
CN108023834A (en) | A kind of cloud resource auto-allocation method and device | |
CN102917014A (en) | Resource scheduling method and device | |
CN109697105A (en) | A kind of container cloud environment physical machine selection method and its system, virtual resource configuration method and moving method | |
CN109948803A (en) | Algorithm model optimization method, device and equipment |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |