1. Embedded System Lab. 최 길 모최 길 모 Kilmo Choi rlfah926@naver.com A Software Memory Partition Approach for Eliminating Bank-level Interference in Multicore Systems Lei Liu, Zehan Cui, Mingjie Xing, Yungang Bao, Mingyu Chen, Chengyong Wu

2. Embedded System Lab. 최 길 모최 길 모 Contents Background and Motivation Bank-Level Partition Mechanism(BPM) Results Conclusion Reference

3. Embedded System Lab. 최 길 모최 길 모 Background and Motivation Memory bank  The same set of memory access speed Multicore platform

4. Embedded System Lab. 최 길 모최 길 모 Background and Motivation Bank-Level Parallelism(BLP) and Bank Sharing  Multiple banks can serve memory requests concurrently and independently  Memory system usually employs a bank-interleaved address mapping schema Memory interference on multicore platform  Causes performance degradation(throughput slowdown and unfairness )  ex. row buffer hit rate decrease from 1 core(over 60%) to 16 core(35%) CoreCore MC CoreCore Bank row buffer conflict row buffer conflict

5. Embedded System Lab. 최 길 모최 길 모 Background and Motivation Numerous new memory scheduling algorithms have been proposed to address the interference problem  However, these algorithms usually employ complex scheduling logic and need hardware modification to memory controllers Bank-level conflicts can be fully eliminated by exclusively mapping a thread’s data to specific banks  How much influence the performance of thread amount of available bank?

6. Embedded System Lab. 최 길 모최 길 모 Bank-Level Partition Mechanism(BPM) Overview of BPM  OS memory management system uses a page-coloring mechanism to partition banks into several groups and maps each thread (process) to a specific bank group  Address mapping policy Advantages  row buffer conflict ↓ row buffer hit ↑  BPM is entirely software approach  Flexible  Easier for OS to monitor thread’s behavior than hardware

7. Embedded System Lab. 최 길 모최 길 모 Bank-Level Partition Mechanism(BPM) Discover bank bits by software method

8. Embedded System Lab. 최 길 모최 길 모 Results Environments  4 cores, 2.8GHz Intel Core i7-860 processor, 8GB DDR3 main memory  CentOS Linux 5.4 with kernel 2.6.32.15  SPEC CPU2006

9. Embedded System Lab. 최 길 모최 길 모 Results Overall system performance

10. Embedded System Lab. 최 길 모최 길 모 Results Page-Policy and Power

11. Embedded System Lab. 최 길 모최 길 모 Results BPM VS Cache-Partition-Only The correlation between BPM improvements and Per-core bandwidth

12. Embedded System Lab. 최 길 모최 길 모 Reference J. Lin, Q. Lu, X. Ding, Z. Zhang, X. Zhang, and P. Sadayappan. Gaining Insights into Multicore Cache Partitioning: Bridging the Gap between Simulation and Real Systems. In HPCA-14, 2008. Dimitris Kaseridis, Jeffrey Stuecheli, Lizy Kurian John. Minimalist Open-page: A DRAM Page-mode Scheduling Policy for the Many- core Era. In MICRO 44, 2011