ETD system

Electronic theses and dissertations repository

 

Tesi etd-04162009-145440


Thesis type
Tesi di dottorato di ricerca
Author
SOLINAS, MARCO
URN
etd-04162009-145440
Title
Cache Architectures for Wire-Delay Dominated CMP Systems
Settore scientifico disciplinare
ING-INF/05
Corso di studi
INGEGNERIA DELL'INFORMAZIONE
Commissione
Relatore Prof. Dini, Gianluca
Relatore Ing. Foglia, Pierfrancesco
Relatore Prof. Prete, Cosimo Antonio
Parole chiave
  • wire-delay
  • NUCA
  • coherence protocol
  • CMP
  • cache memory
Data inizio appello
29/05/2009;
Consultabilità
parziale
Data di rilascio
29/05/2049
Riassunto analitico
Increasing on-chip wire delay and growing off-chip miss latency, present two key<br>challenges in designing large Level-2 (L2) CMP caches. Currently, some CMPs<br>use a shared L2 cache to maximize cache capacity and minimize off-chip misses.<br>Others use private L2 caches, replicating data to limit the delay from slow on-chip<br>wires and minimize cache access time. Ideally, to improve performance for a wide<br>variety of workloads, CMPs prefer both the capacity of a shared cache and the<br>access latency of private caches. In this context, NUCA caches have been proved<br>to be able to tolerate wire delay effects while maintaining a huge on-chip storage<br>capacity.<br>In this thesis, we investigate the choice of the coherence strategy (MESI and<br>MOESI) and the whole system topology as design tradeoffs for S-NUCA based<br>CMP system, and propose and evaluate a novel block migration scheme for DNUCA<br>based systems, in which are addressed two specific problems that can arise<br>due to the presence of multiple traffic sources.<br>Results show that, in S-NUCA based CMP systems, choosing between MESI and<br>MOESI has not a significant impact on performance, while the system topology can<br>lead to very different behaviors.<br>Block migration is introduced in NUCA cache to reduce access latency in a shared<br>cache. Our results show that the migration mechanism is effective in reducing the<br>average L1 miss latency, but the impact on performance is smaller, as a<br>consequence of the very little L1 miss rate.
File