MadCache: A PC-aware Cache Insertion Policy
نویسندگان
چکیده
While the field of computer architecture is always looking for novel research directions to bring improved performance and efficiency, it is often simple improvements to more mature topics that have the most substantial impact. Cache replacement policy is one such research area, where innovations are highly sought after because of their direct improvement on performance. Furthermore, as chip-multiprocessors have become the dominant chip design, new cache replacement schemes should seek to improve the performance of workloads for both singlethreaded and shared cache multithreaded systems. In this paper we propose MadCache, a cache insertion policy that uses memory access history based on the Program Counter (PC) to determine the appropriate policy for the L3 cache. A PC-based history table stores information regarding cache accesses and determines whether the L3 should default to the LRU replacement policy for workloads that exhibit good locality or bypass for streaming memory accesses. Furthermore, this PC-based history table allows individual PCs to override this default policy if its history indicates a behavior significantly different from the typical trend of the workload. We show that MadCache is able to improve IPC by 2.5% over LRU for a singlethreaded 1MB 16-way L3 cache. Finally, we extend MadCache to a four thread, 4MB shared L3 cache and demonstrate a 6% improvement in throughput and 4.5% speedup over LRU averaged across the mixed benchmarks we tested.
منابع مشابه
Runtime-Assisted Shared Cache Insertion Policies Based on Re-reference Intervals
Processor speed is improving at a faster rate than the speed of main memory, which makes memory accesses increasingly expensive. One way to solve this problem is to reduce miss ratio of the processor’s last level cache by improving its replacement policy. We approach the problem by co-designing the runtime system and hardware and exploiting the semantics of the applications written in data-flow...
متن کاملCache Replacement Policy Using Map-based Adaptive Insertion
In this paper, we propose a map-based adaptive insertion policy (MAIP) for a novel cache replacement. The MAIP estimates the data reuse possibility on the basis of data reuse history. To track data reuse history, the MAIP employs a bitmap data structure, which we call memory access map. The memory access map holds all memory accessed locations in a fixed sized memory area to detect the data reu...
متن کاملLLC Dead Block Prediction Considered Not Useful
Dead block predictors (DBPs) improve cache efficiency by identifying blocks that have exhausted their useful lifetime in the cache and prioritizing them for eviction regardless of how much reuse these blocks have experienced. In doing so, DBPs go beyond insertion policies that target only those blocks that have no cache reuse. But how much value do DBPs add over insertion policies? This work ex...
متن کاملThe Commonly Used Lru Replacement Policy Causes Thrashing for Memory- Intensive Workloads. a Simple Mechanism That Dynamically Changes the Insertion Policy Used by Lru Replacement Reduces Cache Misses by 21 Percent and Requires a Total Storage Overhead of Less Than
......One of the major limiters of computer system performance has been the access to main memory, which is typically two orders of magnitude slower than the processor. To bridge this gap, modern processors already devote more than half the on-chip transistors to the last-level cache (in our studies, the L2 cache). These designs typically use the least-recently-used (LRU) replacement policy, or...
متن کامل