A probabilistic model of memory acceses for efficient CPU caching
Abstract/Contents
- Abstract
- The memory hierarchy in a modern processor -- the main memory and different cache levels, consumes about half of the full system energy. We reduce the memory hierarchy energy by three methods: a) reducing main memory accesses by building better last level cache replacement policies, b) reducing wire energy in LLCs by doing intelligent placement and movement of data, and, c) reducing L1 access energy by performing accurate assignment of cache ways to memory access instructions. To design these policies, we propose a new model for program access patterns: the IID Sequence Model (ISM), which uses the reuse distance distributions of different cache lines. We show that a) our proposed Probabilistic Replacement Policy (PRP) reduces LLC misses by 6.6% over state-of-the-art replacement policies such as SHiP [91], b) Sub-Level Insertion Policy (SLIP), which places and moves cache lines according to their reuse distance distribution, reduces L2 energy by 35% and L3 energy by 22%, and, c) Cache Way Assignment (CWA) reduces L1D cache energy consumption by 14% by reducing way mispredictions.
Description
Type of resource | text |
---|---|
Form | electronic; electronic resource; remote |
Extent | 1 online resource. |
Publication date | 2016 |
Issuance | monographic |
Language | English |
Creators/Contributors
Associated with | Das, Subhasis |
---|---|
Associated with | Stanford University, Department of Electrical Engineering. |
Primary advisor | Dally, William J |
Thesis advisor | Dally, William J |
Thesis advisor | Aamodt, Tor M |
Thesis advisor | Rosenblum, Mendel |
Advisor | Aamodt, Tor M |
Advisor | Rosenblum, Mendel |
Subjects
Genre | Theses |
---|
Bibliographic information
Statement of responsibility | Subhasis Das. |
---|---|
Note | Submitted to the Department of Electrical Engineering. |
Thesis | Thesis (Ph.D.)--Stanford University, 2016. |
Location | electronic resource |
Access conditions
- Copyright
- © 2016 by Subhasis Das
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...