CN117170852A - Computing power distribution, service and test method, system and storage medium - Google Patents
Computing power distribution, service and test method, system and storage medium Download PDFInfo
- Publication number
- CN117170852A CN117170852A CN202210575050.7A CN202210575050A CN117170852A CN 117170852 A CN117170852 A CN 117170852A CN 202210575050 A CN202210575050 A CN 202210575050A CN 117170852 A CN117170852 A CN 117170852A
- Authority
- CN
- China
- Prior art keywords
- level
- service
- determining
- computing power
- request
- 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.)
- Pending
Links
- 238000009826 distribution Methods 0.000 title claims abstract description 98
- 238000003860 storage Methods 0.000 title claims abstract description 13
- 238000010998 test method Methods 0.000 title claims description 6
- 238000000034 method Methods 0.000 claims abstract description 74
- 230000000694 effects Effects 0.000 claims abstract description 59
- 238000012545 processing Methods 0.000 claims abstract description 53
- 238000012360 testing method Methods 0.000 claims abstract description 14
- 238000004364 calculation method Methods 0.000 claims description 61
- 230000001105 regulatory effect Effects 0.000 claims description 56
- 230000003631 expected effect Effects 0.000 claims description 33
- 230000008569 process Effects 0.000 claims description 26
- 230000033228 biological regulation Effects 0.000 claims description 25
- 238000011156 evaluation Methods 0.000 claims description 24
- 230000008901 benefit Effects 0.000 claims description 13
- 230000008859 change Effects 0.000 claims description 13
- 238000004590 computer program Methods 0.000 claims description 10
- 238000012937 correction Methods 0.000 claims description 6
- 238000002474 experimental method Methods 0.000 claims description 5
- 230000001186 cumulative effect Effects 0.000 claims 1
- 238000005516 engineering process Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 16
- 230000006870 function Effects 0.000 description 5
- 102000037983 regulatory factors Human genes 0.000 description 3
- 108091008025 regulatory factors Proteins 0.000 description 3
- 230000001960 triggered effect Effects 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/22—Detection or location of defective computer hardware by testing during standby operation or during idle time, e.g. start-up testing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/48—Program initiating; Program switching, e.g. by interrupt
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
Abstract
This disclosure proposes a method for computing power allocation, service, and testing, as well as a system and storage medium, related to the field of computer technology. The computing power distribution method comprises the following steps: determining relative gain of an effect of processing a current service request relative to consumption for each computing power distribution level; and comparing the relative gains corresponding to different computing power distribution levels, and determining the computing power distribution level with the maximum relative gain as the computing power distribution level of service equipment for processing the service request. Through the method, the effective utilization rate of the computing power can be improved.
Description
Technical Field
The present disclosure relates to the field of computer technology, and in particular, to a computing power distribution, service and testing method, and a system and storage medium.
Background
The online service receives the request in real time, and the result is returned in real time through the internal processing of the service. The benefits brought by processing different requests are different, and the occupied calculation power is different. Computational effort is a machine resource that deploys online services and can be considered the cost of processing requests.
Disclosure of Invention
The inventors noted that: the related art can adopt a mode that logic of each request processing is the same and unified degradation is performed after the upper limit of the machine computing power is exceeded. However, the logic for handling each request in this way is the same, and the low QPS (query-per-second) machines have low computational power requirements, and the high QPS machines have high computational power requirements, and the number of machines needed needs to be distributed based on the daily peak QPS, resulting in low computational power utilization of the machines for most of the time in the day.
It is an object of the present disclosure to improve the effective utilization of the computing power of a device.
According to an aspect of some embodiments of the present disclosure, there is provided a computing force distribution method, comprising: assigning a level to each computing power, determining a relative gain of an effect of processing the current service request relative to consumption; and comparing the relative gains corresponding to the different calculation power distribution levels, and determining the calculation power distribution level with the maximum relative gain as the calculation power distribution level of the service equipment for processing the service request.
In some embodiments, determining the relative gain of the effect of processing the current service request with respect to consumption comprises: acquiring the calculation power consumption parameter of the service equipment; determining an expected effect evaluation value for processing the current service request; determining a real-time regulation factor according to the use state error of the service equipment and the historical regulation factor or according to the effect gain rate brought by adjusting the calculation power distribution level under the current QPS condition, and updating the historical regulation factor by using the real-time regulation factor; and determining the relative gain of the effect relative to the consumption according to the calculated power consumption parameter, the regulation factor and the expected effect evaluation value of the service equipment.
In some embodiments, determining the real-time regulatory factor based on the usage status error of the service device and the historical regulatory factor, or based on the relative gain over a predetermined length of time from the current time instant, comprises: determining the change rate of the current QPS; comparing the change rate of the QPS with a preset threshold value, and if the change rate of the QPS is smaller than or equal to the preset threshold value, determining a first regulating factor according to the use state error of the service equipment and the historical regulating factor to serve as a real-time regulating factor; if the change rate of the QPS is larger than a preset threshold, determining a second regulating factor according to the effect gain rate brought by adjusting the calculation power distribution level under the current QPS condition, and taking the second regulating factor as a real-time regulating factor.
In some embodiments, determining the first regulatory factor based on the usage status error of the service device and the historical regulatory factor comprises: determining one or more usage state parameters and corresponding target state parameters of the service device; determining a correction proportion based on a proportional integral derivative PID according to the use state parameter and the target state parameter; and regulating the historical regulating factor according to the corrected proportion to obtain a first regulating factor.
In some embodiments, the usage status parameters include at least one of CPU utilization, memory usage, request time consumption, request failure rate.
In some embodiments, determining the second regulatory factor based on the effective gain rate that would result from adjusting the calculated force allocation level in the current QPS case comprises: according to the current QPS, determining a feasible calculation power distribution level based on historical data as a basic level; determining a benefit gain rate using a second level compared to a third level based on the base level, wherein the second level is a calculated force allocation level one level higher than the base level, and the third level is a calculated force allocation level one level lower than the base level; and taking the benefit gain rate as a second regulating factor.
In some embodiments, determining the benefit-gain-ratio for the second level compared to the third level based on the base level comprises: determining the sum of expected effect estimated values of all service requests in one second under the condition of adopting a second level to process the current QPS as a second level effect; determining the sum of expected effect estimated values of all service requests in one second under the condition of adopting a third level to process the current QPS as a third level effect; determining the accumulated calculation power consumption gap of calculation power consumption parameters of the second level and the third level under the condition of processing all service requests within one second under the current QPS condition; and determining the benefit gain rate according to the second level effect, the third level effect and the accumulated calculated force consumption gap.
In some embodiments, determining the benefit-gain-ratio for the second level compared to the third level based on the base level comprises: the benefit gain rate λ is determined according to the following formula:
wherein i is a request identifier; j is a basic level, j+1 is a second level, and j-1 is a third level; q (Q) ij+1 To employ a second level of expected effectiveness assessment for request i, Q ij-1 Adopting a third level of expected effect evaluation value for the request i; q j+1 For the second level of the power consumption parameter, q j-1 Calculating a power consumption parameter for a third level; n is QPS.
In some embodiments, determining the relative gain of the effect of processing the current service request with respect to consumption comprises: according to formula Q j -λq j Determining relative gain, wherein j is the calculated force allocation level for which Q j Evaluation of the average expected effect of individual service requests with the use of the calculation effort assignment rank j, q j The calculation force consumption parameter in the case of assigning the rank j to the adoption of the calculation force.
According to an aspect of some embodiments of the present disclosure, there is provided a service method, including: the service equipment receives a service request; determining a computing force allocation level according to any of the computing force distribution methods mentioned above; the service device processes the service request according to the power distribution level.
According to an aspect of some embodiments of the present disclosure, a test method is presented, comprising: sending the request of the comparison group to the first service equipment for processing; sending the experiment group request to a second service device for processing; the experimental result is determined by comparing the processing results of the control group request and the experimental group request, wherein the calculation power distribution level of the first service device and the second service device for each service request is determined according to any one of the calculation power distribution methods mentioned above.
According to an aspect of some embodiments of the present disclosure, there is provided a computing force distribution device comprising: a relative gain determining unit configured to assign a level to each computing power, determine a relative gain of an effect of processing the service request with respect to consumption; and the computing force distribution determining unit is configured to compare the relative gains corresponding to different computing force distribution levels and determine the computing force distribution level with the largest relative gain.
In some embodiments, the relative gain determination unit comprises: a consumption parameter acquisition subunit configured to allocate a level for each computing power, to acquire a computing power consumption parameter of the service device; an expected effect determination subunit configured to determine, for each computing power allocation level, an expected effect evaluation value for processing the current service request; a regulation factor determining subunit configured to determine, for each calculation power allocation level, a real-time regulation factor according to a use state error of the service taking device and a history regulation factor, or according to an effect gain rate that would be brought by adjusting the calculation power allocation level in the case of the current QPS, and update the history regulation factor with the real-time regulation factor; and a relative gain determining subunit configured to assign a level to each computing power, and determine a relative gain of the effect with respect to the consumption based on the computing power consumption parameter, the regulation factor, and the expected effect evaluation value of the service taking device.
According to an aspect of some embodiments of the present disclosure, there is provided a computing force distribution device comprising: a memory; and a processor coupled to the memory, the processor configured to perform any of the computing force allocation methods mentioned above based on instructions stored in the memory.
According to an aspect of some embodiments of the present disclosure, a non-transitory computer-readable storage medium is presented, having stored thereon computer program instructions which, when executed by a processor, implement the steps of any of the computing force allocation methods mentioned above.
According to an aspect of some embodiments of the present disclosure, there is provided a service system comprising: any of the computing force distribution devices mentioned above; and one or more service devices configured to receive the service request and process the service request according to the computing power allocation level determined by the computing power allocation apparatus.
According to an aspect of some embodiments of the present disclosure, there is provided a test system comprising: a first service device configured to process a control group request for the test; a second service device configured to process the experimental group request for testing; and any of the computing power distribution means mentioned hereinabove, configured to determine a computing power distribution level for the first service device and the second service device to process each service request.
Drawings
The accompanying drawings, which are included to provide a further understanding of the disclosure, illustrate and explain the present disclosure, and together with the description serve to explain the present disclosure. In the drawings:
FIG. 1 is a flow chart of some embodiments of the computing force distribution method of the present disclosure
Fig. 2A is a flow chart of other embodiments of the computing force distribution method of the present disclosure.
FIG. 2B is a flow chart of some embodiments of determining real-time regulatory factors in the computational force distribution method of the present disclosure.
Fig. 3 is a flow chart of some embodiments of the service method of the present disclosure.
Fig. 4 is a flow chart of some embodiments of the test method of the present disclosure.
Fig. 5A is a schematic diagram of some embodiments of a computing force distribution device of the present disclosure.
Fig. 5B is a schematic diagram of some embodiments of a portion of a computing force distribution device of the present disclosure.
Fig. 6 is a schematic diagram of further embodiments of the computing force distribution device of the present disclosure.
Fig. 7 is a schematic view of yet other embodiments of the computing force distribution device of the present disclosure.
Fig. 8 is a schematic diagram of some embodiments of a service system of the present disclosure.
Fig. 9 is a schematic diagram of some embodiments of a system architecture of a service system of the present disclosure.
Fig. 10 is a schematic diagram of some embodiments of a test system of the present disclosure.
Detailed Description
The technical scheme of the present disclosure is described in further detail below through the accompanying drawings and examples.
A flow chart of some embodiments of the computing force distribution method of the present disclosure is shown in fig. 1.
In step 11, a level is assigned for each computing power, and a relative gain of the effect of processing the current service request with respect to consumption is determined. In some embodiments, the operation in step 11 may be triggered each time a service request is received by the service device. In some embodiments, the consumption may include consumption of resources such as a CPU, and may also refer to the time consumed in processing a single service request. The effect of the service request may be determined from a preset or statistical determination for historical data.
In step 12, the relative gains corresponding to the different computing power allocation levels are compared, and the computing power allocation level with the largest relative gain is determined as the computing power allocation level of the service device for processing the service request.
Based on the mode in the embodiment, the corresponding calculation force distribution level can be determined according to the service request from two angles of consumption and effect, so that the calculation force utilization rate of the equipment is improved, the calculation force consumption is enabled to bring higher effect, and the effective utilization rate of the calculation force is improved, in consideration of the characteristics of the request and the equipment performance.
In some embodiments, the relative gain of the effect of processing a service request with respect to consumption may vary depending on the computing power allocation level employed and the variation of the service request. A flow chart of some embodiments of the computing force distribution method of the present disclosure is shown in fig. 2A.
First, in the case where the service device receives a service request, the operations in steps 210 to 240 are performed for each calculation power allocation level for the service device.
In step 210, the power consumption parameters of the service device are obtained. In some embodiments, the power consumption parameter is the average time taken by the service device to process a service request. In some embodiments, the power consumption parameter q of the service device at each power distribution level may be determined by statistical operations on the historical data j Where j is the power consumption level identification.
In step 220, an expected effectiveness assessment value for processing the current service request is determined. In some embodiments, the expected effectiveness assessment of the service request received by the current service device may be determined by statistical operations on the historical data. In some embodiments, the expected effect evaluation value may be a preset value corresponding to the service request, and the preset value may be related to the type and importance of the service request. In some embodiments, the expected effect evaluation value may also be related to the computing power distribution level, and the expected effect evaluation value Q may be determined according to the corresponding relationship between the preset value and the service request i, the computing power distribution level ij 。
In some embodiments, one or two models of the calculated power consumption parameter and the expected effect evaluation value may also be determined in advance according to the historical behavior data training, and further in steps 110 and 120, the calculated power consumption parameter and the expected effect evaluation value are obtained in real time by running the relevant models.
In step 230, real-time regulatory factors are determined based on service equipment usage status errors and historical regulatory factors, or based on the effective gain rate that would be achieved by adjusting the computing power allocation level in the current QPS case. The history regulatory factor may be a real-time regulatory factor determined in the previous execution process, and the history regulatory factor is updated after each determination of the real-time regulatory factor.
In some embodiments, a scheme of determining the real-time regulatory factor according to the service equipment usage state error and the historical regulatory factor may be adopted in the case that the QPS is relatively stable; under the condition that the QPS fluctuation is large, a scheme of determining the real-time regulation factor according to the effect gain rate brought by adjusting the calculation power distribution level under the current QPS condition is adopted.
In some embodiments, the service device usage status error includes an error between a target value and an actual value of a CPU (Central Processing Unit ) utilization, memory usage, request time consumption, request failure rate, etc. of the service device.
In some embodiments, a possible computing power distribution level under the current QPS may be determined in advance according to a computing power consumption parameter of the service device, and further, based on the computing power distribution level, an adjusted effect gain rate for making a predetermined level difference may be determined, where the effect gain rate may be solved according to a relationship curve between the effect and the computing power distribution level. In some embodiments, the concept of benefit gain rate may be defined by the following equation (1):
wherein j1 and j2 are respectively assigned with grade marks for calculation force. In some embodiments, j2=j1-1. Q (Q) j Mean value, q of expected effect evaluation values for each service request under the power distribution level j j The calculation force consumption parameters under the level j are allocated to the calculation force.
In step 240, a relative gain of the effect relative to the consumption is determined based on the calculated power consumption parameter, the regulatory factor, and the expected effect assessment value for the service device.
In some embodiments, the formula Q may be followed j -λq j Determining relative gain, wherein j is the calculated force allocation level for which Q j Assigning an average of expected effectiveness estimates at level j, i.e., an average expected effectiveness estimate for a single service request, q j The computing power consumption parameters of the device under the class j are allocated to the computing power.
After obtaining the relative gain for each of the power distribution levels, step 250 is performed.
In step 250, the relative gains corresponding to the different power distribution levels are compared, and the power distribution level with the greatest relative gain is determined as the power distribution level of the service device for processing the service request, i.e. max (Q j -λq j ) Corresponding toTo determine a computing power distribution level. The computing power allocation level is to be the computing power allocation level at which the service device is to service the current service request.
In some embodiments, the computing force allocation level may be determined based on a backpack problem. The calculation force distribution is processed by adopting a knapsack problem, and the following formula group is obtained:
wherein, C represents calculation force constraint, namely, one second of machine calculation force, such as CPU-1 second;
q j a computational effort consumption parameter representing action-j;
Q ij representing the request quality, namely, the estimated income of the request i by adopting action-j;
x ij for a coefficient having a value of 0 or 1.
Therefore, the knapsack problem of the calculation power distribution can be understood as that all requests are processed in unit time under the premise of performance constraint, and the effect is maximized. Performance constraints refer to CPU duty cycle, tp99 (the minimum time required to satisfy ninety-nine percent of network requests) time consuming to satisfy the respective upper threshold constraint.
Solving the knapsack problem can take the knapsack problem as a conditional extremum problem into a Lagrange formula to obtain:
s.t.λ≥0
μ i ≥0
x ij ≥0
lambda and mu in the formula are Lagrangian factors, and the solution of the condition extremum is obtained as follows:
q j for the calculation of the force consumption parameter λ is the effect gain per unit time. Each request only needs to be processed by solving for max (Q ij -λq j ) Can determine max (Q ij -λq j ) And j is the corresponding j value, and j is the calculation power distribution grade, so that the optimization of the whole effect is realized.
Based on the mode in the embodiment, the calculation force consumption can bring higher effect, and the effective utilization rate of the calculation force is improved; by generating and adjusting the real-time regulation factors, the accuracy and objectivity of the relative gain can be improved, and the effective utilization rate is further improved.
For step 230 described above, a flow chart of some embodiments of the computing force distribution method of the present disclosure is shown in fig. 2B.
In step 231, the rate of change of the current QPS is determined. In some embodiments, the QPS may be determined by collecting parameters such as the amount of service request, time, etc., received by the service device.
In step 232, it is determined whether the rate of change of QPS is equal to or less than a predetermined threshold. If the rate of change of the QPS is less than or equal to the predetermined threshold, then step 233 is performed; otherwise, step 234 is performed.
In some embodiments, the predetermined threshold may be set, adjusted, e.g., set to any value between 20 and 40, such as 30, depending on the requirements and test.
In step 233, under the current situation, it is determined that the QPS is relatively stable, no substantial fluctuation occurs, and the first regulatory factor may be determined according to the usage status error of the service device and the historical regulatory factor, and used as the real-time regulatory factor.
In some embodiments, one or more usage state parameters of the service device and corresponding target state parameters are first determined, where the usage state parameters may include one or more of CPU utilization, memory usage, request time consumption, and request failure rate. Taking the request time as an example, the usage state parameter is the actual time consumption of processing the request, and the corresponding target state parameter is the expected time consumption of processing the request by using the history regulatory factor.
In some embodiments, the correction ratio may be determined based on a PID algorithm according to the usage state parameter and the target state parameter, and the historical regulator may be adjusted according to the correction ratio to obtain the first regulator. Specifically, the correction ratio u (t) is determined according to the formula (2), and the first regulating factor lambda is determined according to the formula (3) 1 ：
λ 1 ＝λ old *(1+8(t)) (3)
Wherein e (t) is an error function based on the usage state parameter and the target state parameter, and may be a difference value between the usage state parameter and the target state parameter; k (K) p 、K i 、K d The coefficients of the ratio, the integral and the derivative respectively. Lambda (lambda) old Is a history regulatory factor.
In some embodiments, the usage status parameters include one or more of CPU utilization, memory usage, request time consumption, request failure rate.
In step 234, a second regulatory factor is determined as a real-time regulatory factor based on the effective gain rate that would result from adjusting the power distribution level in the current QPS scenario.
In some embodiments, a feasible computing power allocation level may be determined as a base level based on historical data according to the current QPS, for example, performing a solving operation according to equation (4):
determining a computing power distribution level j satisfying the condition of formula (4) asA base level, where q j The computing power consumption parameters in the case of assigning a class j to the computing power, such as the average time consumed per processing of a service request; n is QPS, the number of service requests received in one second.
And further determining, based on the base level, that the benefit gain rate of the second level is used as a second regulation factor compared with that of a third level, wherein the second level is marked as j+1, and the third level is marked as j-1.
Specifically, in some embodiments, the sum of the estimated values of the expected effects of all service requests in one second in the case of processing the current QPS with the second level may be determined first as the second level effect; determining the sum of expected effect estimated values of all service requests in one second under the condition of adopting a third level to process the current QPS as a third level effect; further determining an accumulated calculation power consumption gap of calculation power consumption parameters of the second level and the third level under the condition of processing all service requests within one second under the current QPS condition; and determining the benefit gain rate according to the second level effect, the third level effect and the accumulated calculated force consumption gap.
The benefit gain rate lambda can be determined according to equation (5) 2 ：
Wherein i is a request identifier; j is a basic level, j+1 is a second level, and j-1 is a third level; q (Q) ij+1 To employ a second level of expected effectiveness assessment for request i, Q ij-1 Adopting a third level of expected effect evaluation value for the request i; q j+1 For the second level of the power consumption parameter, q j-1 Calculating a power consumption parameter for a third level; n is QPS, the number of service requests received in one second.
Based on the mode in the embodiment, the sudden change of the flow can be timely reflected while the stable service request flow is responded, and the reliability of the elastic calculation force distribution is improved, so that the effective utilization rate of the calculation force of the service equipment is further improved.
A flowchart of some embodiments of the service method of the present disclosure is shown in fig. 3.
In step 310, the service device receives a service request. In some embodiments, execution of step 320 may be triggered each time a service request is received by a service device.
In step 320, a computing force distribution level is determined according to any of the computing force distribution methods mentioned above.
In step 330, the service device allocates corresponding computing power resources according to the computing power allocation level determined in step 320, to process the service request received in step 310.
Based on the method in the embodiment, the calculation force resources can be elastically allocated in real time according to the online request, so that the calculation force utilization rate of the equipment is improved, the calculation force consumption is brought to a higher effect, and the calculation force is more effectively utilized in consideration of the characteristics of the request and the equipment performance.
In some embodiments, the service method described above may be applied in an ABTest scenario, and a flowchart of some embodiments of the test method of the present disclosure is shown in fig. 4.
In step 410, a control group request is sent to the first service device for processing. In some embodiments, the first service device may be a single service machine or may be a service group.
In step 420, the experiment set request is sent to a second service device for processing. In some embodiments, the second service device may be a single service machine or may be a service group.
The computing power allocation level of each service request by the first service device and the second service device is determined according to any one of the computing power allocation methods mentioned above, thereby realizing the elastic computing power allocation for each request.
In step 430, the experimental results are determined by comparing the results of the treatment of the control group request and the experimental group request, such as determining whether the experimental group has produced an effect gain over the control group.
In the related art, different request processing logics are adjusted according to the performance indexes of the real-time machines, and the difference exists between the performances of different machines at the same time, so that the same request has different logics adopted on different machines, and the experimental effect is affected. Based on the manner in the above embodiments of the present disclosure, the confidence level of the ABTest experiment effect can be improved because the same real-time adjustment factor is adopted for different service devices in the execution process of the computing power distribution method.
A schematic diagram of some embodiments of the computing force distribution device of the present disclosure is shown in fig. 5A.
The relative gain determination unit 51 is capable of determining, for each computing power allocation level, a relative gain of the effect of processing the current service request with respect to consumption.
The computing power allocation determination unit 52 can determine the computing power allocation level by comparing the relative gains corresponding to the different computing power allocation levels after obtaining the relative gain of each computing power allocation level, and determining the computing power allocation level with the largest relative gain as the computing power allocation level for the service device to process the service request. The computing power allocation level is to be the computing power allocation level at which the service device is to service the current service request.
The device can trigger from two aspects of consumption and effect, and determines the corresponding computing power distribution level aiming at the service request, so that the computing power utilization rate of the equipment is improved, the computing power consumption is enabled to bring higher effect, and the effective utilization rate of computing power is improved, in consideration of the characteristics of the request and the equipment performance.
In some embodiments, the relative gain determination unit 51 is as shown in fig. 5B.
The consumption parameter acquisition sub-unit 511 can allocate a level for each computing power, acquire the computing power consumption parameter of the service device. In some embodiments, the power consumption parameter is the average time taken by the service device to process a service request. In some embodiments, the power consumption parameter q of the service device at each power distribution level may be determined by statistical operations on the historical data j Where j is the power consumption level identification.
The expected effect determination subunit 512 is capable of assigning a level to each computing force, determining that the process is currentAn expected effectiveness evaluation value of the service request. In some embodiments, the expected effect evaluation value may be a preset value corresponding to the service request, and the preset value may be related to the type and importance of the service request. In some embodiments, the expected effect evaluation value may also be related to the computing power distribution level, and the expected effect evaluation value Q may be determined according to the corresponding relationship between the preset value and the service request i, the computing power distribution level ij 。
The regulation factor determination subunit 513 can determine the real-time regulation factor for each calculation power allocation level based on the use state error of the service device and the history regulation factor, or based on the effect gain rate that would be brought about by raising the calculation power allocation level in the case of the current QPS. The history regulatory factor may be a real-time regulatory factor determined in the previous execution process, and the history regulatory factor is updated after each determination of the real-time regulatory factor.
The relative gain determination subunit 514 is capable of assigning a level to each computing power, and determining a relative gain of the effect with respect to the consumption based on the computing power consumption parameter, the regulation factor, and the expected effect evaluation value of the service device.
The device can bring higher effect to the calculation force consumption and improve the effective utilization rate of the calculation force; by generating and adjusting the real-time regulation factors, the accuracy and objectivity of the relative gain can be improved, and the effective utilization rate is further improved.
In some embodiments, the regulatory factor determining subunit 513 may determine the real-time regulatory factor according to the manner in the embodiment shown in fig. 2B, so that the abrupt change of the flow can be timely reflected while the smooth service request flow is handled, and the reliability of the elastic computing force distribution is improved, so that the effective utilization rate of the computing force of the service device is further improved.
A schematic structural diagram of one embodiment of the computing force distribution device of the present disclosure is shown in fig. 6. The computing force distribution device comprises a memory 601 and a processor 602. Wherein: the memory 601 may be a magnetic disk, flash memory or any other non-volatile storage medium. The memory is used to store instructions in the corresponding embodiments of the computing force allocation method hereinabove. The processor 602 is coupled to the memory 601 and may be implemented as one or more integrated circuits, such as a microprocessor or microcontroller. The processor 602 is configured to execute instructions stored in the memory, so that the computing power consumption can bring higher effects while improving the computing power utilization rate of the device, and the effective utilization rate of the computing power can be improved.
In one embodiment, the computing force distribution device 700 may also include a memory 701 and a processor 702, as shown in FIG. 7. The processor 702 is coupled to the memory 701 through a BUS 703. The computing power distribution device 700 may also be coupled to external storage device 705 via storage interface 704 for invoking external data, and may also be coupled to a network or another computer system (not shown) via network interface 706. And will not be described in detail herein.
In this embodiment, the data instruction is stored in the memory, and the processor processes the instruction, so that the computing power utilization rate of the device can be improved, and meanwhile, the computing power consumption can bring a higher effect, and the effective utilization rate of the computing power can be improved.
In another embodiment, a computer readable storage medium has stored thereon computer program instructions which, when executed by a processor, implement the steps of the method of the corresponding embodiment of the computing force distribution method. It will be apparent to those skilled in the art that embodiments of the present disclosure may be provided as a method, apparatus, or computer program product. Accordingly, the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present disclosure may take the form of a computer program product embodied on one or more computer-usable non-transitory storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
A schematic diagram of some embodiments of the service system of the present disclosure is shown in fig. 8.
The service system includes any one of the calculation force distribution devices 81 mentioned above, and the service apparatuses 821 to 82n, n being a positive integer of 1 or more.
The service device is capable of receiving the service request and processing the service request according to the computing power allocation level determined by the computing power allocation means. In some embodiments, the service device may trigger the computing power allocation device 81 in real time to determine a computing power allocation level after receiving the service request, and further determine computing power resources allocated for the service request according to the computing power allocation level.
The service system can flexibly allocate the computing power resources according to the online request in real time, so that the computing power consumption is brought to a higher effect while the computing power utilization rate of the equipment is improved in consideration of the characteristics of the request and the equipment performance, and the computing power is more effectively utilized.
In some embodiments, the architecture of the service system of the present disclosure may be as shown in fig. 9.
The evaluation service module 92 can estimate the expected effect estimate Q for each request with different computing power allocation levels action-j j And the calculated power consumption parameters q of different actions-j j 。
The regulatory backbone module 93 is capable of adjusting the real-time regulatory factor 941 based on the QPS, flow distribution, and current machine performance, and feeding back to the on-line service module's elastic system SDK (Software Development Kit ) unit 911.
The evaluation service module 92 and the regulatory backbone module 93 may perform automated data processing for data connections with the various service devices of the online service module 91 in relatively independent devices and servers, determining and feeding back the regulatory factor 941 to the online service module. In some embodiments, the assessment service module may run a model capable of determining the calculated force consumption parameter and the expected effect estimate, thereby determining the calculated force consumption parameter and the expected effect estimate in real-time.
The online service module 91 includes a service device and an elastic system SDK unit 911.
The SDK unit 911 of the elastic system can regulate the factor, Q according to the real time j And q j And calculating the action-j with the maximum relative gain of each request in real time, and further executing computing power distribution by the service equipment according to the action-j. In some embodiments, in terms of hardware, the resilient system SDK units may be integrated into eachAnd in the online service module, solving the action-j calculation force distribution optimum adopted at the current moment of each request in real time. In some embodiments, the resilient system SDK unit is also capable of providing immediate feedback based on sudden increases or decreases in flow, thereby directing the distribution of the flow's computing power.
The service system can adopt the unit integrated on the online service module to carry out subsequent operation after the real-time regulation and control factors are determined based on the deployment mode, thereby improving the timeliness of the feedback of the calculation power distribution level and improving the processing efficiency.
A schematic diagram of some embodiments of the test system of the present disclosure is shown in fig. 10.
The first service device 1021 is able to process a control group request for testing. In some embodiments, the first service device may be a single service machine or may be a service group.
The second service 1022 is capable of handling requests for experimental groups for testing. In some embodiments, the second service device may be a single service machine or may be a service group.
The computing power distribution apparatus 1010 is capable of determining a computing power distribution level for each service request processed by the first service device and the second service device.
In some embodiments, after the request is processed, the experimental results can be determined by comparing the processing results of the control group request and the experimental group request, such as determining whether the experimental group produces an effect gain over the control group. In some embodiments, the comparison operation may be performed manually, increasing flexibility; the comparison device can also be arranged for comparison, so that the processing efficiency is improved.
According to the test system, the same real-time adjustment factors are adopted for different service devices in the execution process of the calculation force distribution method, so that the confidence level of the ABtest experiment effect can be improved.
The present disclosure is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the disclosure. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Thus far, the present disclosure has been described in detail. In order to avoid obscuring the concepts of the present disclosure, some details known in the art are not described. How to implement the solutions disclosed herein will be fully apparent to those skilled in the art from the above description.
The methods and apparatus of the present disclosure may be implemented in a number of ways. For example, the methods and apparatus of the present disclosure may be implemented by software, hardware, firmware, or any combination of software, hardware, firmware. The above-described sequence of steps for the method is for illustration only, and the steps of the method of the present disclosure are not limited to the sequence specifically described above unless specifically stated otherwise. Furthermore, in some embodiments, the present disclosure may also be implemented as programs recorded in a recording medium, the programs including machine-readable instructions for implementing the methods according to the present disclosure. Thus, the present disclosure also covers a recording medium storing a program for executing the method according to the present disclosure.
Finally, it should be noted that: the above embodiments are merely for illustrating the technical solution of the present disclosure and are not limiting thereof; although the present disclosure has been described in detail with reference to preferred embodiments, those of ordinary skill in the art will appreciate that: modifications may be made to the specific embodiments of the disclosure or equivalents may be substituted for part of the technical features; without departing from the spirit of the technical solutions of the present disclosure, it should be covered in the scope of the technical solutions claimed in the present disclosure.
Claims (16)
1. A method of computing force distribution, comprising:
assigning a level to each computing power, determining a relative gain of an effect of processing the current service request relative to consumption;
and comparing the relative gains corresponding to the different calculation power distribution levels, and determining the calculation power distribution level with the maximum relative gain as the calculation power distribution level of the service equipment for processing the service request.
2. The method of claim 1, wherein the determining a relative gain of an effect of processing the current service request relative to consumption comprises:
acquiring the calculation power consumption parameter of the service equipment;
determining an expected effect evaluation value for processing the current service request;
determining a real-time regulation factor according to a service equipment use state error and a history regulation factor or according to an effect gain rate brought by adjusting a calculation power distribution level under the current query rate QPS condition, and updating the history regulation factor by using the real-time regulation factor;
and determining the relative gain of the effect relative to the consumption according to the calculated power consumption parameter, the regulating factor and the expected effect evaluation value of the service equipment.
3. The method of claim 2, wherein the determining the real-time regulatory factor based on the service device usage status error and the historical regulatory factor, or based on the relative gain for a predetermined length of time up to the current time, comprises:
determining the change rate of the current QPS;
comparing the rate of change of the QPS to a predetermined threshold,
if the change rate of the QPS is smaller than or equal to the preset threshold value, determining a first regulating factor according to the use state error of the service equipment and the historical regulating factor, and taking the first regulating factor as the real-time regulating factor;
and if the change rate of the QPS is larger than the preset threshold, determining a second regulating factor as the real-time regulating factor according to the effect gain rate brought by adjusting the calculation power distribution level under the current QPS condition.
4. The method of claim 3, wherein the determining the first regulatory factor based on the service device usage status error and the historical regulatory factor comprises:
determining one or more usage state parameters and corresponding target state parameters of the service device;
determining a correction proportion based on a proportional integral derivative PID according to the use state parameter and the target state parameter;
and regulating the historical regulating factor according to the correction proportion to obtain the first regulating factor.
5. The method of claim 4, wherein the usage status parameters include at least one of CPU utilization, memory footprint, request time consumption, request failure rate.
6. A method according to claim 3, wherein said determining a second regulatory factor based on an effect gain rate that would result from adjusting the calculated force allocation level in the current QPS case comprises:
according to the current QPS, determining a feasible calculation power distribution level based on historical data as a basic level;
determining a benefit gain rate using a second level compared to a third level based on the base level, wherein the second level is a higher computing power allocation level than the base level and the third level is a lower computing power allocation level than the base level;
and taking the benefit gain rate as the second regulating factor.
7. The method of claim 6, wherein the determining, based on the base level, a benefit-gain ratio for the second level compared to the third level comprises:
determining the sum of expected effect estimated values of all service requests in one second under the condition of adopting the second level to process the current QPS as a second level effect;
determining the sum of expected effect estimated values of all service requests in one second under the current QPS condition processed by the third level as a third level effect;
determining an accumulated computing power consumption difference between the computing power consumption parameter of the second level and the computing power consumption parameter of the third level under the condition of processing all service requests within one second under the current QPS condition;
and determining the benefit gain rate according to the second level effect, the third level effect and the cumulative calculation force consumption gap.
8. The method of claim 6, wherein the determining, based on the base level, a benefit-gain ratio for the second level compared to the third level comprises:
the benefit gain rate λ is determined according to the following formula:
wherein i is a request identifier; j is the basic level, j+1 is the second level, and j-1 is the third level; q (Q) ij+1 To employ the second level of expected effectiveness assessment value for request i, Q ij-1 Adopting the third level of expected effect evaluation value for request i; q j+1 For the second level of the calculated power consumption parameter, q j-1 A power consumption parameter for the third level; n is QPS.
9. The method of claim 2, wherein the determining a relative gain of the effect of processing the current service request relative to consumption comprises:
according to formula Q j -λq j Determining relative gain, wherein j is the calculated force allocation level for which Q j Evaluation of the average expected effect of individual service requests with the use of the calculation effort assignment rank j, q j For the calculation power consumption parameter under the condition of adopting calculation power distribution grade j, lambda is a real-time regulation factor.
10. A method of service, comprising:
the service equipment receives a service request;
the computing power distribution method according to any one of claims 1 to 9, determining a computing power distribution level;
the service device processes the service request according to the computing power distribution level.
11. A method of testing, comprising:
sending the request of the comparison group to the first service equipment for processing;
sending the experiment group request to a second service device for processing;
determining an experimental result by comparing the processing results of the control group request and the experimental group request,
wherein the computing power allocation level of each service request by the first service device and the second service device is determined according to the method of any one of claims 1-9.
12. A computing force distribution device, comprising:
a relative gain determination unit configured to assign a level to each computing power, determine a relative gain of an effect of processing the service request with respect to consumption; and
and the computing power distribution determining unit is configured to compare the relative gains corresponding to different computing power distribution levels and determine the computing power distribution level with the largest relative gain as the computing power distribution level of the service equipment for processing the service request.
13. A computing force distribution device, comprising:
a memory; and
a processor coupled to the memory, the processor configured to perform the method of any of claims 1-9 based on instructions stored in the memory.
14. A non-transitory computer readable storage medium having stored thereon computer program instructions which, when executed by a processor, implement the steps of the method of any of claims 1 to 9.
15. A service system, comprising:
the computing force distribution device of claim 12 or 13; and
one or more service devices configured to receive a service request and process the service request according to the computing power allocation level determined by the computing power allocation apparatus.
16. A test system, comprising:
a first service device configured to process a control group request for the test;
a second service device configured to process the experimental group request for testing; and
the computing power allocation apparatus of claim 12 or 13, configured to determine a computing power allocation level at which the first service device and the second service device process each service request.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210575050.7A CN117170852A (en) | 2022-05-25 | 2022-05-25 | Computing power distribution, service and test method, system and storage medium |
| PCT/CN2023/081654 WO2023226545A1 (en) | 2022-05-25 | 2023-03-15 | Computing power distribution method, computing power service method, computing power testing method, and system and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210575050.7A CN117170852A (en) | 2022-05-25 | 2022-05-25 | Computing power distribution, service and test method, system and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117170852A true CN117170852A (en) | 2023-12-05 |
Family
ID=88918368
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210575050.7A Pending CN117170852A (en) | 2022-05-25 | 2022-05-25 | Computing power distribution, service and test method, system and storage medium |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN117170852A (en) |
| WO (1) | WO2023226545A1 (en) |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113448721A (en) * | 2020-03-27 | 2021-09-28 | 中国移动通信有限公司研究院 | Network system for computing power processing and computing power processing method |
| CN111679905B (en) * | 2020-05-11 | 2022-03-08 | 天津大学 | Calculation network fusion network model system |
| CN111507650B (en) * | 2020-07-02 | 2021-01-05 | 深圳微品致远信息科技有限公司 | Computing power distribution scheduling method and system for edge computing platform |
| CN114518940A (en) * | 2020-11-19 | 2022-05-20 | 北京希姆计算科技有限公司 | Task scheduling circuit, method, electronic device and computer-readable storage medium |
| CN112286644B (en) * | 2020-12-25 | 2021-05-28 | 同盾控股有限公司 | Elastic scheduling method, system, equipment and storage medium for GPU (graphics processing Unit) virtualization computing power |
-
2022
- 2022-05-25 CN CN202210575050.7A patent/CN117170852A/en active Pending
-
2023
- 2023-03-15 WO PCT/CN2023/081654 patent/WO2023226545A1/en unknown
Also Published As
| Publication number | Publication date |
|---|---|
| WO2023226545A1 (en) | 2023-11-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10761897B2 (en) | Predictive model-based intelligent system for automatically scaling and managing provisioned computing resources | |
| US11886919B2 (en) | Directing queries to nodes of a cluster of a container orchestration platform distributed across a host system and a hardware accelerator of the host system | |
| JP2018036779A (en) | Electronic device, method, and information processing system | |
| CN108205469B (en) | MapReduce-based resource allocation method and server | |
| Mondal et al. | Scheduling of time-varying workloads using reinforcement learning | |
| CN110231984B (en) | Multi-workflow task allocation method and device, computer equipment and storage medium | |
| CN112231085A (en) | Mobile terminal task migration method based on time perception in collaborative environment | |
| US10066604B2 (en) | Method and system for hybrid wind power generation | |
| CN115543624A (en) | Heterogeneous computing power arrangement scheduling method, system, equipment and storage medium | |
| CN113296905A (en) | Scheduling method, scheduling device, electronic equipment, storage medium and software product | |
| CN108241526B (en) | Adaptive control strategy based on mode prediction | |
| CN117170852A (en) | Computing power distribution, service and test method, system and storage medium | |
| CN116596286B (en) | Optimized scheduling method, device and equipment for virtual power plant and storage medium | |
| CN111125097B (en) | Report scheduling method and device | |
| US11513866B1 (en) | Method and system for managing resource utilization based on reinforcement learning | |
| US11169990B2 (en) | Re-computing pre-computed search results | |
| CN113535346B (en) | Method, device, equipment and computer storage medium for adjusting thread number | |
| CN113326132A (en) | Information adjusting method, equipment and storage medium | |
| US7970884B1 (en) | Distribution of intermediate data in a multistage computer application | |
| CN115344388B (en) | Power real-time simulation calculation task allocation method and device considering communication and calculation force | |
| US20230325764A1 (en) | Optimization of maximum quantity allowed per large order | |
| WO2019116083A1 (en) | Dynamic adjustment of workload forecast | |
| CN116703165B (en) | Electric power metering data security risk assessment method and device | |
| Pace et al. | Dynamic Resource Shaping for Compute Clusters | |
| US20240095081A1 (en) | Optimizing operation of high-performance computing systems |
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 |