2021/02/23 by Nadja Ramhöj Holtryd, Holtryd, Nadja Ramhöj, Madhavan Manivannan +5
Computer Science · #Cloud Computing and Resource Management #Distributed systems and fault tolerance #FOS: Computer and information sciences #Hardware Architecture (cs.AR) #Parallel Computing and Optimization Techniques
paper · pdf · doi:10.48550/arxiv.2102.11528
openalex publication_date 2021/02/23 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Reducing the average memory access time is crucial for improving the\nperformance of applications running on multi-core architectures. With workload\nconsolidation this becomes increasingly challenging due to shared resource\ncontention. Techniques for partitioning of shared resources - cache and\nbandwidth - and prefetching throttling have been proposed to mitigate\ncontention and reduce the average memory access time. However, existing\nproposals only employ a single or a subset of these techniques and are\ntherefore not able to exploit the full potential of coordinated management of\ncache, bandwidth and prefetching. Our characterization results show that\napplication performance, in several cases, is sensitive to prefetching, cache\nand bandwidth allocation. Furthermore, the results show that managing these\ntogether provides higher performance potential during workload consolidation as\nit enables more resource trade-offs. In this paper, we propose CBP a\ncoordination mechanism for dynamically managing prefetching throttling, cache\nand bandwidth partitioning, in order to reduce average memory access time and\nimprove performance. CBP works by employing individual resource managers to\ndetermine the appropriate setting for each resource and a coordinating\nmechanism in order to enable inter-resource trade-offs. Our evaluation on a\n16-core CMP shows that CBP, on average, improves performance by 11% compared to\nthe state-of-the-art technique that manages cache partitioning and prefetching\nand by 50% compared to the baseline without cache partitioning, bandwidth\npartitioning and prefetch throttling.\n