CN104516784A - Method and system for forecasting task resource waiting time - Google Patents
Method and system for forecasting task resource waiting time Download PDFInfo
- Publication number
- CN104516784A CN104516784A CN201410796248.3A CN201410796248A CN104516784A CN 104516784 A CN104516784 A CN 104516784A CN 201410796248 A CN201410796248 A CN 201410796248A CN 104516784 A CN104516784 A CN 104516784A
- Authority
- CN
- China
- Prior art keywords
- task
- task record
- period
- record
- historic
- 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
Abstract
The invention discloses a method and a system for forecasting task resource waiting time and relates to the large-scale calculating system resource management, optimization and distribution, in particular to the method and system for forecasting the task resource waiting time. The method for forecasting the task resource waiting time includes that acquiring historical task records, deleting task records with dependency relationships from the historical task records, and generating new historical task records; acquiring task records in the time period related to the time period to be forecast from the new historical task records through self-correlation functions to generate a task record set; setting a task resource waiting time threshold, acquiring the number of the task records of which the task resource waiting time is longer than the task resource waiting time threshold from the task record set, and forecasting the task resource waiting time in the time period to be forecast through a bayesian method according to the total amount of the task records in the task record set. The method and system for forecasting the task resource waiting time are capable of forecasting the usability of the calculating system resource and optimizing the task dispatching.
Description
Technical field
The present invention relates to large-scale computing systems (comprising supercomputing and cloud computing) resource management, optimize and distribute, particularly one prediction task resource stand-by period method and system.
Background technology
Portraying of using for large-scale computing systems resource, prediction, optimizes, and in the process of distributing, existing many schemes adopt the method based on model.Specifically, first researchist chooses one or more and uses relevant dimension to carry out the data tracking observation (remaining runtime of such as operation with system resource, the queuing time of operation in system queue, etc.), then certain dependent probability model is applied to portray the probability distribution of this kind of dimension data, next, researchist applies the prediction that probability distribution character that this kind of model present is carried out for this system future performance, thus realize resource optimization and reasonable distribution, such as, Downey application logarithm is uniformly distributed (log uniform distribution) and measures operation remaining runtime, the concept (BinomialMethod Batch Predictor (BMBP)) that Brevik and co-worker thereof propose binomial method batch forecast carrys out describing system queue wait time, Li and co-worker's application mix Gauss model (Gaussian-mixture model) thereof describe distribution working time of the whole operation of system in certain section of special time.
Can be there are some following practical problemss in the method for portraying resource distribution based on particular probability model: the historic load record of mass data is difficult to obey certain specific probability distribution in actual applications in the middle of the practical application processing large-scale computing system management and large data processing, in fact, by checking different real data, certain single probability distribution not only disobeyed in the historic load record of large data, and even not easy-to-use mixing probability Distribution Model is portrayed; In the operation property hypothesis reality independent of each other submitted in short time interval about user in some conventional probability models (such as binomial distribution model (binomialmodel) and derivative model thereof) often and be false, in fact, by the concrete research to actual supercomputer historic load, we find that most of user can repeatedly submit to content similar at short notice, the operation that parameter is suitable, therefore carries out the Evaluation and Prediction of resource consumption with this type of probability model inaccurate often.
Summary of the invention
For the deficiencies in the prior art, the present invention proposes a kind of prediction task resource stand-by period method and system.
The present invention proposes a kind of method predicting the task resource stand-by period, comprising:
Step 1, obtains historic task record, deletes the task record of Existence dependency relationship in this historic task record, generates new historical task record;
Step 2, by autocorrelation function, obtains in this new historical task record the task record had with the time period to be predicted in the time period of correlativity, generates task record set;
Step 3, task resource stand-by period threshold value is set, obtain the number that the task resource stand-by period in this task record set exceedes the task record of this task resource stand-by period threshold value, and according to the total amount of task record in this task record set, predict the task resource stand-by period in this time period to be predicted by bayes method.
The method of described prediction task resource stand-by period, the concrete steps of this step 1 comprise:
Step 11, judges task record in this historic task record whether Existence dependency relationship;
Step 12, if exist, deletes the task record of Existence dependency relationship.
The method of described prediction task resource stand-by period, this step 11 comprises:
Step 21, select time critical point t* and space critical point x*;
Step 22, if to be interposed between the submission time of two task records in this historic task record within time critical point t* and the degree of closing on of parameter choose within space interval x*, and the density of pairing is higher than the density of pairing except these two task records in this historic task record, then these two task records have dependence within (t*, x*) yardstick.
The method of described prediction task resource stand-by period, this step 12 comprises:
Step 31, arranges the task record in this historic task record by submission time is ascending;
Step 32, from the task record that submission time is minimum, the Selecting parameter degree of closing on of deleting this task record all in this minimum task record of this submission time to following t* time and that this submission time is minimum is less than the task record of x*;
Step 33, upgrades this historic task record and repeats this step 22, until travel through this historic task record, upgrades this historic task record.
The method of described prediction task resource stand-by period, in this step 3, the formula of this bayes method is:
Wherein N
lTW, k-1, k-2for this task resource stand-by period exceedes this number of this task record of this task resource stand-by period threshold value, N
k-1, k-2for this total amount of this task record in this task record set, P
lTW, kfor in this time period to be predicted, the task resource stand-by period exceedes the task probability of this task resource stand-by period threshold value.
The present invention also proposes a kind of system predicting the task resource stand-by period, comprising:
Generating new historical task record module, for obtaining historic task record, deleting the task record of Existence dependency relationship in this historic task record, generate new historical task record;
Generate task record set, for by autocorrelation function, obtain in this new historical task record the task record had with the time period to be predicted in the time period of correlativity, generate task record set;
Prediction module, for arranging task resource stand-by period threshold value, obtain the number that the task resource stand-by period in this task record set exceedes the task record of this task resource stand-by period threshold value, and according to the total amount of task record in this task record set, predict the task resource stand-by period in this time period to be predicted by bayes method.
The system of described prediction task resource stand-by period, this generation new historical task record module also comprises:
Judge module, judges task record in this historic task record whether Existence dependency relationship;
Delete dependence module, if exist, delete the task record of Existence dependency relationship.
The system of described prediction task resource stand-by period, the concrete effect of this judge module comprises: select time critical point t* and space critical point x*; If to be interposed between the submission time of two task records in this historic task record within time critical point t* and the degree of closing on of parameter choose within space interval x*, and the density of pairing is higher than the density of pairing except these two task records in this historic task record, then these two task records have dependence within (t*, x*) yardstick.
The system of described prediction task resource stand-by period, the concrete effect of this deletion dependence module comprises: arranged by submission time is ascending by the task record in this historic task record; From the task record that submission time is minimum, the Selecting parameter degree of closing on of deleting this task record all in this minimum task record of this submission time to following t* time and that this submission time is minimum is less than the task record of x*; Upgrade this historic task record and repeat this step 22, until travel through this historic task record, upgrading this historic task record.
The system of described prediction task resource stand-by period, in this prediction module, the formula of this bayes method is:
Wherein N
lTW, k-1, k-2for this task resource stand-by period exceedes this number of this task record of this task resource stand-by period threshold value, N
k-1, k-2for this total amount of this task record in this task record set, P
lTW, kfor in this time period to be predicted, the task resource stand-by period exceedes the task probability of this task resource stand-by period threshold value.
From above scheme, the invention has the advantages that:
The present invention can be used for the workability of prediction and calculation system resource, can be used to optimize job scheduling, optimize the configuration of task resource demand, improve resource utilization, and the prediction of the stand-by period of operation in Job execution queue, experiment of the present invention shows that the present invention can reach the reliable prediction rate of more than 89%.
Accompanying drawing explanation
Fig. 1 is overview flow chart of the present invention;
Fig. 2 is Job partition counting chart;
Fig. 3 waits as long for probability chart after denoising decorrelation;
Fig. 4 is the LTW month trend map on supercomputer siren (Kraken);
Fig. 5 waits as long for probability automatic correlation chart the moon;
Fig. 6 is the process flow diagram that historic task record carries out noise reduction process.
Wherein Reference numeral is:
Step 100 is overall step of the present invention, comprising:
Step 101/102/103;
Step 200 is noise reduction concrete steps of the present invention, comprising:
Step 201/202/203.
Embodiment
Inventor, in the process of the historic load of research supercomputer siren (Kraken), finds that operation (task) stand-by period in Job execution queue that the user of about 52.4% submits to will exceed its actual run time.For each operation, particularly concurrent job, two basic parameters are had to have expressed its demand for calculation resources: the number (CPU, internal memory) of working time, computing node.The present invention is from the angle of user, wish to ensure under the condition precedent obtaining correct result, adjust this two parameters, make operation can be run faster and reduce the stand-by period in Job execution queue, factory work logging is being carried out in the process of statistical study, certain user at short notice repeatedly submit similar operation (such as same executive routine to, different input data etc.) have impact on effective statistics greatly, because the operation number that system can perform for same user generally all can limit to some extent simultaneously.The present invention passes through statistical method: space and temporal clustering detection method (space-time clustering detection method) (also known as Knox method (Knox method)), Task Dependent relation in analytic system load histories record, be different from common cluster (clustering) method, the method that the present invention uses more emphasizes the relation of parameter space and time, thus make data reduction (reduction) more effective, remove the task record having dependence, reach removal noise, load histories record after denoising can be used for generating and wait as long for probability chart accurately, this waits as long for probability could be chosen and wait as long for probability tables (Long-Time Waiting Chart) macroscopical trend in dedicating to different parameters value, and then give user's overall direction about parameter choose.。Further, the present invention can generate simultaneously in units of the moon, wait as long for probability trend map, utilization rate due to computational resource is dynamic, be not that more historical record is more helpful for the calculating of delay probability, often unnecessary data have the negative effect covering true trend, so the present invention uses autocorrelation function AutoCorrelation Function (ACF, autocorrelation function has different portraying in different field, the present invention adopts auto-covariance function to describe coefficient of autocorrelation, auto-covariance function describes time series X (t) at any two not t1 in the same time, second order mixing center square between the value of t2, be used for describing the degree of correlation of X (t) in the fluctuations (relative and average) of two moment values, autocorrelation function also referred to as centralization) determine the time interval region with correlative connection, on test macro, such as find that the data of 3 months have relation associated with each other, finally, the present invention uses bayes method (BayesianFramework) to predict, and Future direction (according to its requirement for system resource) obtains resource and the stand-by period run on this computing system.
Below in conjunction with accompanying drawing, the specific embodiment of the present invention is described.
Be below general steps of the present invention, as shown in Figure 1, concrete steps are as follows:
Step 101, the Task Dependent relation of (historic task record) in usage space and temporal clustering detection method (space-time clusteringdetection method) (also known as Knox method) analytic system load histories record, remove the task record having dependence, reach removal noise, improve the object of prediction accuracy.
Step 102, probability chart (Long-Time Waiting Chart is waited as long for according to load histories record (new historical task record) generation after removing noise, task resource waits as long for chart), generate simultaneously and wait as long for probability trend map in units of the moon.In addition, autocorrelation function AutoCorrelation Function (ACF) is used to determine the time interval region with correlative connection.
Step 103, Future direction (according to its requirement for system resource) obtains resource and the stand-by period run on a computing system to use bayes method (Bayesian Framework) to predict, task resource stand-by period threshold value is wherein set, obtain the quantity that this task resource stand-by period in this task record set exceedes the task record of this task resource stand-by period threshold value, and according to the total quantity of task record in this task record set, predict the task resource stand-by period in this certain time period by bayes method.
Below in conjunction with embodiment, the present invention is further described:
Excavate and wait as long for pattern really: Job partition count table: first, statistics is obtained (for the operation that each is submitted to the form of two-dimensional diagram to the resource service condition of operation, particularly concurrent job, two basic parameters are had to have expressed the demand of operation for calculation resources: reserved working time (WallClock Requested (WCR)), number (the CPU of reserved computing node, internal memory) (Number of computeNodes Requested (NNR)), the present embodiment only have employed above two parameters, i.e. reserved working time and reserved computing node number, therefore adopt two-dimentional icon shows statistics, the situation of multidimensional by that analogy), each Parameter Subsection is added up, in an experiment, the work data on the supercomputer siren (Kraken) obtained, (NNR is in the interval of 1 to 4128 node, WCR was in the interval of 0 to 24 hour) be divided into several parameter subregions, the selection of subregion critical value is mainly based on following 2 points: comprise parameter combinations mode that some practical operation user habits select (such as 1 hour working time and 10 computing nodes in each region, 12 hour working time and 1 computing node etc.), ensure to cover sizable charge book in each subregion, thus ensure reliability and the validity of statistics.Fig. 2 is the Job partition count table of the historic load generation according to 30 months on supercomputer siren (Kraken).The noise reduction decorrelation process of historical data: by space and temporal clustering detection method (space-time clustering detectionmethod), also known as Knox method (Knox method), Task Dependent relation in checking system load histories record, concrete grammar is as follows: diacritical point t* and space critical point x* when first selecting according to priori, if to be interposed between the submission time of two operations within time critical point t* and the degree of closing on of parameter choose within space interval x*, then think that two operations are considered to there is relevant possibility, if the density that the density of this type of operation pairing is matched far above other operations, then think that historical data is at (t*, x*) there is very strong correlativity within yardstick, thus need to take noise reduction decorrelation process, in order to check above-mentioned phenomenon, the present invention is applied in the Knox Statistical Identifying Method of application in epidemic transmission, if Knox method display historical data is at (t*, x*) there is very strong correlativity within yardstick, then apply following algorithm to carry out historic load (historic task record) noise reduction process, as shown in Figure 6:
Step 201, historic load is pressed the arrangement of submission time ascending order, perform step 202, from first load data, all and this load data Selecting parameter degree of closing in from this load data submission time to the following t* time of deleting is less than all related load data of x*, step 203, upgrades load data and repeats step 202, until travel through all historic load.
The present invention uses above algorithm to carry out noise reduction decorrelation process to the load histories of system, thus reaches minimizing data volume, produces relatively independent charge book.
Wait as long for probability chart: invention defines and wait as long for (LTW:Long TimeWaiting) threshold value (task resource stand-by period threshold value), such as 1 hour, based on the new data after noise reduction and removal correlativity, set up form as shown in Figure 2, the ratio that operation that stand-by period is greater than LTW threshold value accounts for All Jobs number in this subregions all is calculated for each subregion, thus produces the heating power chart (heatmap) as Fig. 3.
Prediction the task resource stand-by period: as Fig. 3 generate noise reduction and decorrelation after wait as long for probability chart can provide macroscopic view parameters Choice Criteria.For siren's supercomputer, under equal service compute amount (service unit), the parameter choose scheme of reserved a small amount of computing time and a large amount of computing node can produce less waiting as long for than the scheme of reserved a large amount of computing time and a small amount of computing node, thus produce supercomputer user resources utilization ratio more efficiently, but, in order to provide meticulousr and have ageing users' guidebook, the present invention also needs to consider time factor, and concrete steps are as follows:
Autocorrelation function is utilized to determine to have the time interval region of correlative connection: in the time series record of the LTW probability in units of the moon as shown in Figure 4, the historical data that the value in a certain moment is often adjacent has the relation of strong correlation, and the present invention uses autocorrelation function AutoCorrelation Function (ACF) to determine to have the time interval region of correlative connection.Specifically, the present invention adopts auto-covariance function to describe coefficient of autocorrelation.Auto-covariance function describes time series X (t) at any two not t1 in the same time, second order mixing center square between the value of t2, be used for describing the degree of correlation of X (t) in the fluctuations (relative and average) of two moment values, also referred to as the autocorrelation function of centralization.Its definition is
Wherein E represents expectation value, X
ithe random variable values of representative when t (i).μ
ithe desired value of representative when t (i), X
i+kthe random variable values of representative when t (i+k), μ
i+kthe desired value of representative when t (i+k), σ
2represent variance.
Fig. 5 shows and waits as long for probability auto-correlation function value the moon based on supercomputer siren (Kraken) factory work logging: as can be seen from Figure 5 of that monthly wait as long for probability (LTW probability) and have statistic correlation significantly with bimestrial LTW probability before this, then utilize history segment data correlation time and bayes method prediction subsequent time period wait as long for probability, specifically, choose the data after noise reduction decorrelation in the time interval part with correlative connection obtained in the previous step, what the mean value of application Beta-binomial distribution hierarchical model (Beta-binomial hierarchical model) predicted subsequent time period waits as long for probability.For supercomputer siren, because the autocorrelation function of historical data shows the time that waits as long for of next month and bimestrial data are relevant before, thus show that the mean value of Beta-binomial distribution hierarchical model equals:
Wherein N
lTW, k-1, k-2bimestrially before being wait as long for operation number, N
k-1, k-2bimestrial total operation number before being, P
lTW, kbe current month wait as long for probability.Thus each subregion of next month can be predicted by the present invention wait as long for probability.
The invention allows for a kind of system predicting the task resource stand-by period, comprise as lower module:
Generating new historical task record module, for obtaining historic task record, deleting the task record of Existence dependency relationship in this historic task record, generate new historical task record;
Generate task record set, for by autocorrelation function, obtain in this new historical task record with this sometime section there is the task record in the time period of correlativity, generate task record set;
Prediction module, for arranging task resource stand-by period threshold value, obtain the quantity that this task resource stand-by period in this task record set exceedes the task record of this task resource stand-by period threshold value, and according to the total quantity of task record in this task record set, predict the task resource stand-by period in this certain time period by bayes method.
Judge module, judges task record in this historic task record whether Existence dependency relationship, wherein select time critical point t* and space critical point x*; If to be interposed between the submission time of two task records in this historic task record within time critical point t* and the degree of closing on of parameter choose within space interval x*, and the density of pairing is higher than the density of pairing except these two task records in this historic task record, then these two task records have dependence within (t*, x*) yardstick.
Delete dependence module, if exist, delete the task record of Existence dependency relationship, wherein the task record in this historic task record is arranged by submission time is ascending; From the task record that submission time is minimum, the Selecting parameter degree of closing on of deleting this task record all in this minimum task record of this submission time to following t* time and that this submission time is minimum is less than the task record of x*; Upgrade this historic task record and repeat this step 22, until travel through this historic task record, upgrading this historic task record.
In this prediction module, the formula of this bayes method is:
Wherein N
lTW, k-1, k-2for this task resource stand-by period exceedes this quantity of this task record of this task resource stand-by period threshold value, N
k-1, k-2for this total quantity of this task record in this task record set, P
lTW, kfor the task resource stand-by period in this sometime section exceedes the task probability of this task resource stand-by period threshold value.
Claims (10)
1. predict the method for task resource stand-by period, it is characterized in that, comprising:
Step 1, obtains historic task record, deletes the task record of Existence dependency relationship in this historic task record, generates new historical task record;
Step 2, by autocorrelation function, obtains in this new historical task record the task record had with the time period to be predicted in the time period of correlativity, generates task record set;
Step 3, task resource stand-by period threshold value is set, obtain the number that the task resource stand-by period in this task record set exceedes the task record of this task resource stand-by period threshold value, and according to the total amount of task record in this task record set, predict the task resource stand-by period in this time period to be predicted by bayes method.
2. the method for prediction task resource stand-by period as claimed in claim 1, it is characterized in that, the concrete steps of this step 1 comprise:
Step 11, judges task record in this historic task record whether Existence dependency relationship;
Step 12, if exist, deletes the task record of Existence dependency relationship.
3. the method for prediction task resource stand-by period as claimed in claim 2, it is characterized in that, this step 11 comprises:
Step 21, select time critical point t* and space critical point x*;
Step 22, if to be interposed between the submission time of two task records in this historic task record within time critical point t* and the degree of closing on of parameter choose within space interval x*, and the density of pairing is higher than the density of pairing except these two task records in this historic task record, then these two task records have dependence within (t*, x*) yardstick.
4. the method for prediction task resource stand-by period as claimed in claim 2, it is characterized in that, this step 12 comprises:
Step 31, arranges the task record in this historic task record by submission time is ascending;
Step 32, from the task record that submission time is minimum, the Selecting parameter degree of closing on of deleting this task record all in this minimum task record of this submission time to following t* time and that this submission time is minimum is less than the task record of x*;
Step 33, upgrades this historic task record and repeats this step 22, until travel through this historic task record, upgrades this historic task record.
5. the method for prediction task resource stand-by period as claimed in claim 1, it is characterized in that, in this step 3, the formula of this bayes method is:
Wherein N
lTW, k-1, k-2for this task resource stand-by period exceedes this number of this task record of this task resource stand-by period threshold value, N
k-1, k-2for this total amount of this task record in this task record set, P
lTW, kfor in this time period to be predicted, the task resource stand-by period exceedes the task probability of this task resource stand-by period threshold value.
6. predict the system of task resource stand-by period, it is characterized in that, comprising:
Generating new historical task record module, for obtaining historic task record, deleting the task record of Existence dependency relationship in this historic task record, generate new historical task record;
Generate task record set, for by autocorrelation function, obtain in this new historical task record the task record had with the time period to be predicted in the time period of correlativity, generate task record set;
Prediction module, for arranging task resource stand-by period threshold value, obtain the number that the task resource stand-by period in this task record set exceedes the task record of this task resource stand-by period threshold value, and according to the total amount of task record in this task record set, predict the task resource stand-by period in this time period to be predicted by bayes method.
7. the system of prediction task resource stand-by period as claimed in claim 6, it is characterized in that, this generation new historical task record module also comprises:
Judge module, judges task record in this historic task record whether Existence dependency relationship;
Delete dependence module, if exist, delete the task record of Existence dependency relationship.
8. the system of prediction task resource stand-by period as claimed in claim 7, it is characterized in that, the concrete effect of this judge module comprises: select time critical point t* and space critical point x*; If to be interposed between the submission time of two task records in this historic task record within time critical point t* and the degree of closing on of parameter choose within space interval x*, and the density of pairing is higher than the density of pairing except these two task records in this historic task record, then these two task records have dependence within (t*, x*) yardstick.
9. the system of prediction task resource stand-by period as claimed in claim 7, it is characterized in that, the concrete effect of this deletion dependence module comprises: arranged by submission time is ascending by the task record in this historic task record; From the task record that submission time is minimum, the Selecting parameter degree of closing on of deleting this task record all in this minimum task record of this submission time to following t* time and that this submission time is minimum is less than the task record of x*; Upgrade this historic task record and repeat this step 22, until travel through this historic task record, upgrading this historic task record.
10. the system of prediction task resource stand-by period as claimed in claim 6, it is characterized in that, in this prediction module, the formula of this bayes method is:
Wherein N
lTW, k-1, k-2for this task resource stand-by period exceedes this number of this task record of this task resource stand-by period threshold value, N
k-1, k-2for this total amount of this task record in this task record set, P
lTW, kfor in this time period to be predicted, the task resource stand-by period exceedes the task probability of this task resource stand-by period threshold value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410796248.3A CN104516784B (en) | 2014-07-11 | 2014-12-18 | A kind of method and system for predicting the task resource stand-by period |
Applications Claiming Priority (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410332344 | 2014-07-11 | ||
| CN2014103323442 | 2014-07-11 | ||
| CN201410796248.3A CN104516784B (en) | 2014-07-11 | 2014-12-18 | A kind of method and system for predicting the task resource stand-by period |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104516784A true CN104516784A (en) | 2015-04-15 |
| CN104516784B CN104516784B (en) | 2018-03-30 |
Family
ID=52792138
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410796248.3A Active CN104516784B (en) | 2014-07-11 | 2014-12-18 | A kind of method and system for predicting the task resource stand-by period |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104516784B (en) |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105094986A (en) * | 2015-07-20 | 2015-11-25 | 暨南大学 | Prediction method and device for burst access behaviors oriented to storage system |
| CN106469088A (en) * | 2015-08-21 | 2017-03-01 | 华为技术有限公司 | A kind of I/O request scheduling method and scheduler |
| CN107766150A (en) * | 2017-09-20 | 2018-03-06 | 电子科技大学 | A kind of job scheduling algorithm based on hadoop |
| CN108536528A (en) * | 2018-03-23 | 2018-09-14 | 湖南大学 | Using the extensive network job scheduling method of perception |
| CN108885562A (en) * | 2016-04-15 | 2018-11-23 | 谷歌有限责任公司 | The modular electronic equipment predicted with task in future and ability |
| CN109324875A (en) * | 2018-09-27 | 2019-02-12 | 杭州电子科技大学 | A kind of data center server power managed and optimization method based on intensified learning |
| CN109408236A (en) * | 2018-10-22 | 2019-03-01 | 福建南威软件有限公司 | A kind of task load equalization methods of ETL on cluster |
| CN109636212A (en) * | 2018-12-19 | 2019-04-16 | 中国科学技术大学 | The prediction technique of operation actual run time |
| CN109740813A (en) * | 2018-12-29 | 2019-05-10 | 上海华力微电子有限公司 | Online product batches run the analyzing and predicting method of goods state in wafer manufacture |
| CN109800085A (en) * | 2018-12-24 | 2019-05-24 | 东软集团股份有限公司 | Detection method, device, storage medium and the electronic equipment of resource distribution |
| CN110706171A (en) * | 2019-09-26 | 2020-01-17 | 中国电子科技集团公司第十一研究所 | Image noise reduction method and device |
| CN112269650A (en) * | 2020-12-18 | 2021-01-26 | 北京达佳互联信息技术有限公司 | Task scheduling method and device, electronic equipment and storage medium |
| CN112486687A (en) * | 2020-12-03 | 2021-03-12 | 重庆邮电大学 | Cloud platform workload prediction method based on multitask learning time sequence |
| CN112711615A (en) * | 2019-10-24 | 2021-04-27 | 富士通株式会社 | Information processing apparatus, information processing method, and computer-readable storage medium |
| CN112764915A (en) * | 2020-12-10 | 2021-05-07 | 国家卫星气象中心(国家空间天气监测预警中心) | Resource consumption analysis method of time correlation |
| CN112988361A (en) * | 2021-05-13 | 2021-06-18 | 神威超算(北京)科技有限公司 | Cluster task allocation method and device and computer readable medium |
| CN110706171B (en) * | 2019-09-26 | 2024-04-26 | 中国电子科技集团公司第十一研究所 | Image noise reduction method and device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020087734A1 (en) * | 2000-12-29 | 2002-07-04 | Marshall Donald Brent | System and method for managing dependencies in a component-based system |
| CN101052224A (en) * | 2007-05-15 | 2007-10-10 | 华为技术有限公司 | Method, system and device for determining excited service quality opportunity |
| CN103559333A (en) * | 2013-10-12 | 2014-02-05 | 华南理工大学 | Biological gene sequencing task model building method based on log |
-
2014
- 2014-12-18 CN CN201410796248.3A patent/CN104516784B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020087734A1 (en) * | 2000-12-29 | 2002-07-04 | Marshall Donald Brent | System and method for managing dependencies in a component-based system |
| CN101052224A (en) * | 2007-05-15 | 2007-10-10 | 华为技术有限公司 | Method, system and device for determining excited service quality opportunity |
| CN103559333A (en) * | 2013-10-12 | 2014-02-05 | 华南理工大学 | Biological gene sequencing task model building method based on log |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105094986A (en) * | 2015-07-20 | 2015-11-25 | 暨南大学 | Prediction method and device for burst access behaviors oriented to storage system |
| CN105094986B (en) * | 2015-07-20 | 2019-03-22 | 暨南大学 | A kind of prediction technique and device of the burst access behavior towards storage system |
| CN106469088B (en) * | 2015-08-21 | 2020-04-28 | 华为技术有限公司 | I/O request scheduling method and scheduler |
| CN106469088A (en) * | 2015-08-21 | 2017-03-01 | 华为技术有限公司 | A kind of I/O request scheduling method and scheduler |
| CN108885562A (en) * | 2016-04-15 | 2018-11-23 | 谷歌有限责任公司 | The modular electronic equipment predicted with task in future and ability |
| CN108885562B (en) * | 2016-04-15 | 2020-07-10 | 谷歌有限责任公司 | Modular electronic device with future task and capability prediction |
| CN107766150A (en) * | 2017-09-20 | 2018-03-06 | 电子科技大学 | A kind of job scheduling algorithm based on hadoop |
| CN108536528A (en) * | 2018-03-23 | 2018-09-14 | 湖南大学 | Using the extensive network job scheduling method of perception |
| CN109324875A (en) * | 2018-09-27 | 2019-02-12 | 杭州电子科技大学 | A kind of data center server power managed and optimization method based on intensified learning |
| CN109408236A (en) * | 2018-10-22 | 2019-03-01 | 福建南威软件有限公司 | A kind of task load equalization methods of ETL on cluster |
| CN109636212A (en) * | 2018-12-19 | 2019-04-16 | 中国科学技术大学 | The prediction technique of operation actual run time |
| CN109800085B (en) * | 2018-12-24 | 2021-01-29 | 东软集团股份有限公司 | Resource configuration detection method and device, storage medium and electronic equipment |
| CN109800085A (en) * | 2018-12-24 | 2019-05-24 | 东软集团股份有限公司 | Detection method, device, storage medium and the electronic equipment of resource distribution |
| CN109740813A (en) * | 2018-12-29 | 2019-05-10 | 上海华力微电子有限公司 | Online product batches run the analyzing and predicting method of goods state in wafer manufacture |
| CN109740813B (en) * | 2018-12-29 | 2020-11-24 | 上海华力微电子有限公司 | Analysis and prediction method for on-line product batch running state in wafer manufacturing |
| CN110706171A (en) * | 2019-09-26 | 2020-01-17 | 中国电子科技集团公司第十一研究所 | Image noise reduction method and device |
| CN110706171B (en) * | 2019-09-26 | 2024-04-26 | 中国电子科技集团公司第十一研究所 | Image noise reduction method and device |
| CN112711615A (en) * | 2019-10-24 | 2021-04-27 | 富士通株式会社 | Information processing apparatus, information processing method, and computer-readable storage medium |
| CN112486687A (en) * | 2020-12-03 | 2021-03-12 | 重庆邮电大学 | Cloud platform workload prediction method based on multitask learning time sequence |
| CN112764915A (en) * | 2020-12-10 | 2021-05-07 | 国家卫星气象中心(国家空间天气监测预警中心) | Resource consumption analysis method of time correlation |
| CN112764915B (en) * | 2020-12-10 | 2023-11-14 | 国家卫星气象中心(国家空间天气监测预警中心) | Resource consumption analysis method of time correlation |
| CN112269650A (en) * | 2020-12-18 | 2021-01-26 | 北京达佳互联信息技术有限公司 | Task scheduling method and device, electronic equipment and storage medium |
| CN112988361A (en) * | 2021-05-13 | 2021-06-18 | 神威超算(北京)科技有限公司 | Cluster task allocation method and device and computer readable medium |
| CN112988361B (en) * | 2021-05-13 | 2021-08-20 | 中诚华隆计算机技术有限公司 | Cluster task allocation method and device and computer readable medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104516784B (en) | 2018-03-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104516784A (en) | Method and system for forecasting task resource waiting time | |
| CN110399222B (en) | GPU cluster deep learning task parallelization method and device and electronic equipment | |
| US10325008B2 (en) | Techniques for estimating compound probability distribution by simulating large empirical samples with scalable parallel and distributed processing | |
| Shyam et al. | Virtual resource prediction in cloud environment: a Bayesian approach | |
| Mayer et al. | Predictable low-latency event detection with parallel complex event processing | |
| US10831633B2 (en) | Methods, apparatuses, and systems for workflow run-time prediction in a distributed computing system | |
| JP5313990B2 (en) | Estimating service resource consumption based on response time | |
| US20160328273A1 (en) | Optimizing workloads in a workload placement system | |
| CN110476172A (en) | Neural framework for convolutional neural networks is searched for | |
| US9471375B2 (en) | Resource bottleneck identification for multi-stage workflows processing | |
| EP2541419A1 (en) | Characterizing web workloads for quality of service prediction | |
| Seneviratne et al. | Task profiling model for load profile prediction | |
| US20120221373A1 (en) | Estimating Business Service Responsiveness | |
| EP4123987A1 (en) | Enhanced selection of cloud architecture profiles | |
| CN108205469B (en) | MapReduce-based resource allocation method and server | |
| GB2527788A (en) | Scheduling applications in a clustered computer system | |
| CN109558248B (en) | Method and system for determining resource allocation parameters for ocean mode calculation | |
| CN110618867A (en) | Method and device for predicting resource usage amount | |
| CN102736896A (en) | Run-ahead approximated computations | |
| Tran et al. | Towards a profound analysis of bags-of-tasks in parallel systems and their performance impact | |
| Mahmoudi et al. | Performance modeling of metric-based serverless computing platforms | |
| Gupta et al. | Long range dependence in cloud servers: a statistical analysis based on google workload trace | |
| CN115098257A (en) | Resource scheduling method, device, equipment and storage medium | |
| CN108897673B (en) | System capacity evaluation method and device | |
| Almomani et al. | Selecting a good stochastic system for the large number of alternatives |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20220507 Address after: 100095 room 217, second floor, building 17, Chuangke town community supporting commercial building, Wenquan Town, Haidian District, Beijing Patentee after: Zhongkehai micro (Beijing) Technology Co.,Ltd. Address before: 100190 No. 6 South Road, Zhongguancun Academy of Sciences, Beijing, Haidian District Patentee before: Institute of Computing Technology, Chinese Academy of Sciences |
|
| TR01 | Transfer of patent right |