Streaming of Shared Memory
Tuesday September 13, 2005
Hamerschlag Hall D-210
This is a practice talk for PACT 2005.
Carnegie Mellon University
Coherence misses in shared-memory multiprocessors account for a
substantial fraction of execution time in many scientific and commercial
workloads. Memory streaming provides a promising solution to the
coherence miss bottleneck because it improves memory level parallelism
and lookahead while using on-chip resources efficiently. In this
talk, I show that the order shared data are consumed by one processor
is correlated to the order in which they were produced by another.
I demonstrate how this phenomenon can be exploited to send Store-ORDered
Streams (SORDS) of shared data from producers to consumers, thereby
eliminating coherent read misses.
Tom Wenisch is a fifth year PhD student in computer architecture,
working with Prof. Babak Falsafi. Tom makes databases run faster