CLU: Co-Optimizing Locality and Utility in Thread-Aware Capacity Management for Shared Last Level Caches

Abstract

Abstract-Most chip-multiprocessors nowadays adopt a large shared last-level cache (SLLC). This paper is motivated by our analysis and evaluation of state-of-the-art cache management proposals which reveal a common weakness. That is, the existing alternative replacement policies and cache partitioning schemes, targeted at optimizing either locality or utility of co-scheduled threads, cannot deliver consistently the best performance under a variety of workloads. Therefore, we propose a novel adaptive scheme, called CLU, to interactively co-optimize the locality and utility of co-scheduled threads in thread-aware SLLC capacity management. CLU employs lightweight monitors to dynamically profile the LRU (least recently used) and BIP (bimodal insertion policy) hit curves of individual threads on runtime, enabling the scheme to co-optimize the locality and utility of concurrent threads and thus adapt to more diverse workloads than the existing approaches. We provide results from extensive execution-driven simulation experiments to demonstrate the feasibility and efficacy of CLU over the existing approaches (TADIP, NUCACHE, TA-DRRIP, UCP, and PIPP).

GDP: Using Dataflow Properties to Accurately Estimate Interference-Free Performance at Runtime

“…In addition, a large body of work use ATDs and missminimization within their shared LLC partitioning schemes [8,23,31,34,42,43,44,45]. These are complementary to this work because using MCP would enable them to select partitions based on system performance rather than LLC misses.…”

Using provenance to efficiently improve metadata searching performance in storage systems

“…The second one is that the metadata index tree used by us is vastly different from traditional full-size index trees, we do not store the whole index tree and only store the frequently searched files' closest branches in memory. Recent studies [21,22] have shown that data which have been queried may also be requeried later, which means the query has a temporal locality and spatial locality. Thus we can consume a few memory spaces to store these query hotspots' adjacent index branches to achieve high performance index search.…”

Curve fitting based shared cache partitioning scheme for energy saving

“…In this letter, based on curve fitted energy formulation, we propose a cache partitioning scheme, which assigns cache ways to applications. Curve fitting is utilized in many works about caches, it is used for cache modeling [5] and management [6] and shows good feasibility and accuracy. Our work starts from analyzing the characteristic of cache miss rate, then two types of lines are chosen to fit the curve of cache miss rate according to the cache size.…”