SpeculativeScheduling

Extension of the ScheduleFlow Simulator (v1.0) to allow speculative request times at submission and during backfill. The following figure shows the main architecture of the code:

Teal modules are included in the ScheduleFlow Simulator. Orange modules are implemented in the current code.

The python objects represent an extension of the classic reservation-based HPC schedulers by using speculation to determine the resource requirements of stochastic applications based on their past behavior. Specifically, we augment the existing HPC model by speculatively overwriting the request times (including the initial one and subsequent ones in case of failures) of an application during submission.

The backfilling algorithm is also extended to include speculation by allowing stochastic jobs to be scheduled into smaller backfilling spaces than their requests (temporary overwrite of their requirements).

More details about implementation details as well as examples can be found on the wiki.

Paper

If you use the resources available here in your work, please cite our paper:

Speculative Scheduling Techniques for Stochastic HPC Applications. Ana Gainaru, Guillaume Pallez (Aupy), Hongyang Sun, Padma Raghavan [ICPP 2019] (Link to the technical report)

@inproceedings{Gainaru:2019:SSS:3337821.3337890,
 author = {Gainaru, Ana and Aupy, Guillaume Pallez and Sun, Hongyang and Raghavan, Padma},
 title = {Speculative Scheduling for Stochastic HPC Applications},
 booktitle = {Proceedings of the 48th International Conference on Parallel Processing},
 series = {ICPP 2019},
 year = {2019},
 isbn = {978-1-4503-6295-5},
 location = {Kyoto, Japan},
 pages = {32:1--32:10},
 articleno = {32},
 numpages = {10},
 url = {http://doi.acm.org/10.1145/3337821.3337890},
 doi = {10.1145/3337821.3337890},
 acmid = {3337890},
 publisher = {ACM},
 address = {New York, NY, USA},
 keywords = {HPC runtime, Scheduling algorithm, stochastic applications},
} 

Usage

There are currently 2 scripts that can be used to trigger different scheduling scenarios:

• The run_speculative_submission script tests the speculative deployment of stochastic applications by using past information of jobs to overwrite their request run time
• The run_speculative_backfill script tests the speculative backfilling algorithm for different ratios of small to large jobs