System and method for job-to-queue performance ranking and resource matching
Systems and apparatuses for predicting computer job performance rankings across different computer architectures, configurations, and clusters and methods for making and using the same. The method can include running a new job on a selected computer system and identifying one or more nearby jobs that are nearby the new job. A performance of each of the nearby jobs across the available computer systems can be ranked. The available computer systems preferably include the selected computer system. The method also can estimate a rank of the new job on each of the available computer systems. In various embodiments, these technologies advantageously can be used to optimize the matching of jobs to computer systems.
1. A method for generating estimated performance rankings of a new job across a plurality of available computer systems, comprising:
running the new job on a selected computer system;
identifying one or more nearby jobs that are nearby the new job;
ranking a performance of a predetermined number of the nearby jobs across the available computer systems including the selected computer system; and
estimating a performance rank of the new job on each of the available computer systems.
2. The method of claim 1 , wherein said ranking the performance comprises ranking the performance of all of the nearby jobs across the available computer systems.
3. The method of claim 1 , wherein said estimating the performance rank includes averaging performance rankings for a predetermined number of the nearby jobs and reporting the averaged performance rankings as the estimated performance rank of the new job.
4. The method of claim 1 , wherein said running the new job includes recording job telemetry data for the new job, and wherein said estimating the performance rank is based at least in part on the recorded job telemetry data.
5. The method of claim 4 , wherein said recording the job telemetry data comprises recording the job telemetry data for the new job during said running the new job.
6. The method of claim 4 , further comprising managing said running the new job based at least in part on the recorded job telemetry data.
7. The method of claim 4 , wherein said recording the job telemetry data comprises recording job telemetry data for all jobs that have previously been run on the available computer systems.
8. The method of claim 7 , wherein said recording the job telemetry data for all jobs includes recording the job telemetry data for all jobs previously run on the available computer systems after said running the new job is complete.
9. The method of claim 7 , wherein said recording the job telemetry data for all jobs comprises recording high-resolution job telemetry data for all jobs previously run on the available computer systems, and wherein said identifying the one or more nearby jobs includes identifying the one or more nearby jobs at least in part on the recorded high-resolution job telemetry data.
10. The method of claim 9 , wherein said identifying the one or more nearby jobs includes projecting the recorded high-resolution job telemetry data into a low-dimension subspace.
11. The method of claim 10 , wherein said projecting the recorded high-resolution job telemetry data comprises mathematically regularizing the recorded high-resolution job telemetry data.
12. The method of claim 10 , wherein said projecting the recorded high-resolution job telemetry data into a low-dimension subspace comprises projecting the recorded high-resolution job telemetry data using principal components of truncated singular value decomposition.
13. The method of claim 9 , wherein said identifying the one or more nearby jobs includes extracting one or more statistics from the recorded high-resolution job telemetry data and identifying the one or more nearby jobs at least in part on said extracting.
14. The method of claim 13 , wherein said extracting the one or more statistics includes extracting an average value from the recorded high-resolution job telemetry data, extracting an average value from the recorded high-resolution job telemetry data or a combination thereof.
15. The method of claim 13 , further comprising regularizing the one or more extracted statistics from the recorded high-resolution job telemetry data.
16. The method of claim 15 , wherein said identifying the one or more nearby jobs includes projecting the regularized extracted statistics from the recorded high-resolution job telemetry data into a low-dimension subspace.
17. The method of claim 16 , wherein said identifying the one or more nearby jobs includes identifying the one or more nearby jobs with the low-dimension subspace based upon a distance metric between the nearby jobs and the new job.
18. The method of claim 17 , wherein said identifying the one or more nearby jobs comprises identifying the one or more nearby jobs with the low-dimension subspace based upon a Euclidean distance between the nearby jobs and the new job.
19. A computer program product for generating estimated performance rankings of a new job across a plurality of available computer systems, the computer program product being encoded on one or more non-transitory machine-readable storage media and comprising:
instruction for running the new job on a selected computer system;
instruction for identifying one or more nearby jobs that are nearby the new job;
instruction for ranking a performance of a predetermined number of the nearby jobs across the available computer systems including the selected computer system; and
instruction for estimating a performance rank of the new job on each of the available computer systems.
20. A system for generating estimated performance rankings of a new job across a plurality of available computer systems, comprising:
a selected computer system for running the new job; and
a server system for identifying one or more nearby jobs that are nearby the new job, ranking a performance of a predetermined number of the nearby jobs across the available computer systems including the selected computer system and estimating a performance rank of the new job on each of the available computer systems.