Stage-aware performance modeling for computer cluster sizing

• US 9,891,959 B2
• Filed: 10/30/2015
• Issued: 02/13/2018
• Est. Priority Date: 10/30/2015
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Li, Min, Salapura, Valentina, Tan, Jian, Wang, Yandong, Zhang, Li
• Primary Examiner(s): Wu, Benjamin
• Assistant Examiner(s): CHU JOY, JORGE A
• Application Number:
  US14/928,997
• Publication Number:
  US 20170123855A1
• Time in Patent Office: 837 Days
• CPC Class Codes: G06F 2209/501 Performance criteria
• CPC Class Codes: G06F 9/5055 considering software capabilities, i.e. software resources associated or available to the machine
• CPC Class Codes: G06F 9/5061 Partitioning or combining of resources
• CPC Class Codes: G06N 3/12 using genetic models

Abstract

A method, apparatus, and computer program product for configuring a computer cluster. Job information identifying a data processing job to be performed is received by a processor unit. The data processing job to be performed comprises a plurality of stages. Cluster information identifying a candidate computer cluster is also received by the processor unit. The processor unit identifies stage performance models for modeled stages that are similar to the plurality of stages. The processor unit predicts predicted stage performance times for performing the plurality of stages on the candidate computer cluster using the stage performance models and combines the predicted stage performance times for the plurality of stages to determine a predicted job performance time. The predicted job performance time may be used to configure the computer cluster.

Claims

1. A computer-implemented method of configuring a computer cluster, the computer implemented method comprising:

receiving, by a processor unit, job information identifying a data processing job to be performed, wherein the data processing job to be performed comprises a plurality of stages, and wherein the job information defines characteristics of the plurality of stages that include number of tasks, resource profile, data access pattern, output selectivity, amount of shuffle, resource consumption dynamicity, and data set content sensitivity of respective stages in the plurality of stages of the data processing job;

receiving, by the processor unit, cluster information identifying a candidate computer cluster;

identifying, by the processor unit, stage performance models for corresponding to modeled stages having similar characteristics to the characteristics of plurality of stages that include the number of tasks, resource profile, data access pattern, output selectivity, amount of shuffle, resource consumption dynamicity, and data set content sensitivity of the respective stages in the plurality of stages of the data processing job;

predicting, by the processor unit, stage performance times for performing the plurality of stages on the candidate computer cluster using the stage performance models;

combining, by the processor unit, the predicted stage performance times to determine a predicted job performance time;

using, by the processor unit, the predicted job performance time to configure the candidate computer cluster to perform the data processing job; and

performing, by the candidate computer cluster, the date processing job.

2. The computer-implemented method of claim 1 further comprising:

determining, by the processor unit, whether a job performance model for the data processing job to be performed is available in a performance model database; and

using, by the processor unit, the job performance model to determine the predicted job performance time in response to the processor unit determining that the job performance model for the data processing job to be performed is available in the performance model database.

3. The computer-implemented method of claim 2 further comprising:

responsive to the processor unit determining that the job performance model for the data processing job to be performed is not available in the performance model database, performing, by the processor unit, a test run on the candidate computer cluster using a test dataset to train a resulting job performance model for the data processing job to be performed on the candidate computer cluster.

4. The computer-implemented method of claim 3 further comprising:

identifying, by the processor unit, the job performance model in the performance model database for a modeled data processing job that is similar to the data processing job to be performed; and

identifying, by the processor unit, the stage performance models in the performance model database for the modeled stages having the similar characteristics to the characteristics of the plurality of stages of the data processing job to be performed on the candidate computer cluster.

5. The computer-implemented method of claim 4 further comprising:

identifying, by the processor unit, the job performance model in the performance model database based on a similarity between a number and sequence of stages in the modeled data processing job and the number and sequence of the plurality of stages comprising the data processing job to be performed on the candidate computer cluster; and

identifying, by the processor unit, the stage performance models in the performance model database based on a similarity of identified characteristics between the plurality of stages of the data processing job to be performed and the modeled stages.

6. The computer-implemented method of claim 1, wherein the job information defines a dataset size, and further comprising:

selecting, by the processor unit, a plurality of candidate computer clusters to perform the data processing job based on the dataset size of the data processing job;

determining, by the processor unit, the predicted job performance time to perform the data processing job to be performed for the dataset size for each candidate computer cluster in the plurality of candidate computer clusters;

determining, by the processor unit, an estimated cost to perform the data processing job to be performed for each candidate computer cluster in the plurality of candidate computer clusters using the predicted job performance time for each candidate computer cluster in the plurality of candidate computer clusters; and

using, by the processor unit, the estimated cost to select a configuration of one of the plurality of candidate computer clusters to configure the candidate computer cluster.

7. The computer-implemented method of claim 1, wherein using, by the processor unit, the predicted job performance time to configure the candidate computer cluster to perform the data processing job further comprises:

configuring, by the processor unit, a size of the candidate computer cluster.

8. A computer system for configuring a computer cluster, the computer system comprising:

a bus system;

a storage device connected to the bus system, wherein the storage device stores computer readable program code; and

a processor unit connected to the bus system, wherein the processor unit executes the computer readable program code to;

receive job information identifying a data processing job to be performed and cluster information identifying a candidate computer cluster, wherein the data processing job to be performed comprises a plurality of stages; and

wherein the job information defines characteristics of the plurality of the stages that include the number of tasks, resource profile, data access pattern, output selectivity, amount of shuffle, resource consumption dynamicity, and data set content sensitivity of respective stages in the plurality of stages of the data processing job;

identify stage performance models corresponding to modeled stages having similar characteristics to the characteristics of that plurality of stages include the number of tasks, resource profile, data access pattern, output selectivity, amount of shuffle, resource consumption dynamicity, and data set content sensitivity of the respective stages in the plurality of stages of the data processing job;

predict stage performance times for performing the plurality of stages on the candidate computer cluster using the stage performance models and combine the predicted stage performance times for the plurality of stages to determine a predicted job performance time;

use the predicted job performance time to configure the candidate computer cluster to perform the data processing job; and

perform the data processing job.

9. The computer system of claim 8, wherein the processor unit further executes the computer readable program code to:

determine whether a job performance model for the data processing job to be performed is available in a performance model database; and

use the job performance model for the data processing job to be performed to determine the predicted job performance time in response to a determination that the job performance model for the data processing job to be performed is available in the performance model database.

10. The computer system of claim 9, wherein the processor unit further executes the computer readable program code to:

perform a test run on the candidate computer cluster using a test dataset to train a resulting job performance model for the data processing job to be performed on the candidate computer cluster in response to determining that the job performance model for the data processing job to be performed is not available in the performance model database; and

use the job performance model for the data processing job to be performed on the candidate computer cluster to determine the predicted job performance time.

11. The computer system of claim 8, wherein the processor unit further executes the computer readable program code to:

identify a job performance model in a performance model database for a modeled data processing job that is similar to the data processing job to be performed based on a similarity between a number and sequence of stages in the modeled data processing job and the number and sequence of the plurality of stages comprising the data processing job to be performed on the candidate computer cluster.

12. The computer system of claim 8, wherein the processor unit further executes the computer readable program code to:

identify the stage performance models in a performance model database based on a similarity of characteristics between the plurality of stages of the data processing job to be performed and the modeled stages.

13. The computer system of claim 8, wherein the processor unit further executes the computer readable program code to:

select a plurality of candidate computer clusters to perform the data processing job based on a dataset size of the data processing job;

determine the predicted job performance time to perform the data processing job to be performed for the dataset size for each candidate computer cluster in the plurality of candidate computer clusters; and

determine an estimated cost to perform the data processing job to be performed for each candidate computer cluster in the plurality of candidate computer clusters using the predicted job performance time for each candidate computer cluster in the plurality of candidate computer clusters.

14. The computer system of claim 8, wherein the processor unit further executes the computer readable program code to:

use the predicted job performance time to configure a size of the candidate computer cluster.

15. A computer program product for configuring a computer cluster, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to cause the computer to perform a method comprising:

receiving job information identifying a data processing job to be performed, wherein the data processing job to be performed comprises a plurality of stages, and wherein the job information defines characteristics of the plurality of stages that include number of tasks, resource profile, data access pattern, output selectivity, amount of shuffle, resource consumption dynamicity, and data set content sensitivity of respective stages in the plurality of stages of the data processing job;

receiving cluster information identifying a candidate computer cluster;

identifying stage performance models corresponding to modeled stages having similar characteristics to the characteristics of the plurality of stages that include the number of tasks, resource profile, data access pattern, output selectivity, amount of shuffle, resource consumption dynamicity, and data set content sensitivity of the respective stages in the plurality of stages of the data processing job;

predicting stage performance times for performing the plurality of stages on the candidate computer cluster using the stage performance models;

combining the predicted stage performance times for the plurality of stages to determine a predicted job performance time;

using the predicted job performance time to configure the candidate computer cluster to perform the data processing job; and

performing the data processing job.

16. The computer program product of claim 15 further comprising:

determining whether a job performance model for the data processing job to be performed is available in a performance model database; and

using the job performance model for the data processing job to be performed to determine the predicted job performance time in response to determining that the job performance model for the data processing job to be performed is available in the performance model database.

17. The computer program product of claim 16 further comprising:

responsive to determining that the job performance model for the data processing job to performed is not available in the performance model database,

performing a test run on the candidate computer cluster using a test dataset to train a resulting job performance model for the data processing job to be performed on the candidate computer cluster.

18. The computer program product of claim 17 further comprising:

to identifying the job performance model in the performance model database for a modeled data processing job that is similar to the data processing job to be performed on the candidate computer cluster.

19. The computer program product of claim 18 further comprising:

identifying the stage performance models in the performance model database based on a similarity of characteristics between the plurality of stages of the data processing job to be performed and the modeled stages; and

identifying the job performance model in the performance model database based on a similarity between a number and sequence of stages in the modeled data processing job and the number and sequence of the plurality of stages comprising the data processing job to be performed on the candidate computer cluster.

20. The computer program product of claim 15 further comprising:

selecting a plurality of candidate computer clusters to perform the data processing job based on a dataset size of the data processing job;

determining the predicted job performance time for performing the data processing job to be performed for the dataset size for each candidate computer cluster in the plurality of candidate computer clusters; and

determining an estimated cost to perform the data processing job to be performed for each candidate computer cluster in the plurality of candidate computer clusters using a corresponding predicted job performance time for each candidate computer cluster.