1 / 12

OS Driven Core Selection for HCMP Systems

OS Driven Core Selection for HCMP Systems. Anand Bhatia, Rishkul Kulkarni. Outcomes. O Characterization – based on static parameters - Comparison – Heterogeneity aware vs oblivious scheduler O A functional core selector – OSDCS Module X Study of performance change O : Complete

joie
Télécharger la présentation

OS Driven Core Selection for HCMP Systems

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. OS Driven Core Selection for HCMP Systems Anand Bhatia, RishkulKulkarni

  2. Outcomes O Characterization – based on static parameters - Comparison – Heterogeneity aware vs oblivious scheduler O A functional core selector – OSDCS Module X Study of performance change O : Complete - : Partial X : Pending

  3. Prior Work • Efficient program scheduling for heterogeneous multi-core processors – J. Chen, L.K. John • Static Characteristics • Euclidean distance based signature matching • HASS: A scheduler for Heterogeneous Multicore Systems - Shelepov, D., Fedorova et. al. • DVFS scaling • Process differences • A Unified View of Core Selection and Application Steering in Heterogeneous Chip Multiprocessors – SandeepNavada • Bottleneck counters • Execution migration

  4. Phase 1 - Characterization • Signature Generation • Static Measurements • Dependence Distance (Issue Width) • Stack Distance (Cache Size) • Branch Variability (Branch Predictor Size) • Instructions Per Branch (Experimental) • Projection Functions • Core Exploration

  5. Phase 1 - Results Representative Core Configs • Issue width | Dependence distance • BP Size | Branch variability • D$ Size | Stack Distance • AL Size | IPB IPC on chosen core configs • Independent runs • Useful to determine workload mix for scheduler

  6. 721sim 1 Phase 2 – OSDCS Module 721sim 2 Job Queue 721sim 3 Selector Issue Unit 721sim 4 Simulated Superscaler chip Interface Module User space Kernel space Selector LKM Fig. : Experimental setup used for ODCS Job Queue 721sim 1 Fake System call handler 721sim 2 721sim 3 721sim 4 Actual system call handler • Design Considerations • Design choices • Job Monitoring mechanisms Vs Fig : Booting OS on 721sim Cores

  7. 721sim 1 Phase 2 – OSDCS Module - 2 721sim 2 Job Queue 721sim 3 Issue Unit Selector 721sim 4 Simulated Superscaler chip Interface Module User space Kernel space Selector LKM Fig. : Experimental setup used for ODCS Design detailed. User selector Selector LKM Scheduling Policies.

  8. Phase 2 - Policies Two different types: • Varying Schedule & Dispatch mechanisms • Greedy • Non work-conserving • Weights • Bias logically similar parameters heavily.

  9. Testing Framework • Core Exploration • Signature generation using projection functions • Selecting cores • OSDCS Module • Virtual Machine with minimal Ubuntu • Run scripts

  10. Limitations Insensitive to behavioral changes in workload phases Oblivious to dynamically changing inputs to workloads

  11. Future Work Furthering OSDCS module to incorporate timing model Investigating specialized cores/accelerators. Devising projection functions for more parameters like IPB and LSQ sizing. Implement enhanced scheduling policies such as confidence based weighing.

  12. Backup slide 1: Preliminary results. Caveat 1 : Limited job diversity, Reason: Many jobs assert out on core configurations chosen. Caveat 2 : Synthetic Batch Batch : 5 5 5 1 5 1 5 1 5 1 1 1 {All same , diverse, All same}

More Related