Presentation is loading. Please wait.

Presentation is loading. Please wait.

Embedded System Lab. 정범종 PIPP: Promotion/Insertion Pseudo-Partitioning of Multi-Core Shared Caches Yuejian Xie et al. ACM, 2009.

Similar presentations


Presentation on theme: "Embedded System Lab. 정범종 PIPP: Promotion/Insertion Pseudo-Partitioning of Multi-Core Shared Caches Yuejian Xie et al. ACM, 2009."— Presentation transcript:

1 Embedded System Lab. 정범종 tuckdae@naver.com PIPP: Promotion/Insertion Pseudo-Partitioning of Multi-Core Shared Caches Yuejian Xie et al. ACM, 2009

2 정 범 종정 범 종 Embedded System Lab. Table of contents Abstract Background Reference paper PIPP Evaluation Conclusion Reference

3 정 범 종정 범 종 Embedded System Lab. Abstract Cache management(e.g., LRU) policies can lead to poor performance and fairness when the multiple cores compete for the limited LLC capacity Different memory access patterns can cause cache contention in different ways propose a new cache management approach that combines dynamic insertion and promotion policies benefits of cache partitioning, adaptive insertion, and capacity stealing all with a single mechanism

4 정 범 종정 범 종 Embedded System Lab. Background MRU, LRU, Promotion policies Cache Partitioning  Cache partitioning reduces worst-case execution time for critical tasks, thereby enhancing CPU utilization, especially for multicore applications  Page coloring, UCP

5 정 범 종정 범 종 Embedded System Lab. Reference paper Capacity management  M. K. Qureshi and Y. N. Patt. Utility-Based Cache Partitioning: A Low- Overhead, High-Performance, Runtime Mechanism to Partition Shared Caches (UCP) Dead-Time management  M. K. Qureshi, A. Jaleel, Y. N. Patt, S. C. S. Jr., and J. Emer. Adaptive Insertion Policies for High-Performance Caching. (DIP)  A. Jaleel, W. Hasenplaugh, M. Qureshi, J. Sebot, S. S. Jr.,and J. Emer. Adaptive Insertion Policies for Managing Shared Caches. (TADIP)

6 정 범 종정 범 종 Embedded System Lab. PIPP Basic PIPP  make use of UCP’s utility monitors to compute the target partitions  Dynamic promotion  Dynamic Insertion  steal Stream-Sensitive PIPP

7 정 범 종정 범 종 Embedded System Lab. Evaluation Performance impact of the different cache management techniques for the weighted IPC speedup (Cooperative Cache Partitioning for Chip Multiprocessors)  PIPP consistently outperforms unmanaged LRU by a large margin (19.0% on the harmonic mean), and also outperforms both UCP and TADIP (10.6% and 10.1%, respectively)  Similar results hold for the quad-core case where PIPP is 21.9% better than LRU, 12.1% better than UCP and 17.5% better than TADIP

8 정 범 종정 범 종 Embedded System Lab. Conclusion In this work, we have introduced a single unified technique that can provide the benefits of capacity management, adaptive insertion and inter-core capacity stealing This work opens several future directions for research

9 정 범 종정 범 종 Embedded System Lab. Q & A

10 정 범 종정 범 종 Embedded System Lab. Backup slide

11 정 범 종정 범 종 Embedded System Lab. Evaluation

12 정 범 종정 범 종 Embedded System Lab. Evaluation

13 정 범 종정 범 종 Embedded System Lab.

14 정 범 종정 범 종


Download ppt "Embedded System Lab. 정범종 PIPP: Promotion/Insertion Pseudo-Partitioning of Multi-Core Shared Caches Yuejian Xie et al. ACM, 2009."

Similar presentations


Ads by Google