Characterizing memory system performance for local and remote accesses in high end SMPs, low end SMPs and clusters of SMPs
Ch. Kurmann, Thomas M. Stricker · 1998
and national laboratories our computer architecture group is currently engaged in designing and building a cluster of PCs from off-the-shelf SMPs linked by a commercial Gigabit interconnect (e.g. Dolphin SCI, Myricom Myrinet or Gigabit Ethernet). As computer scientists we are not trying to deploy some cheap GigaFlops to our colleagues in computational chemistry or physics, but are most interested in understanding the performance of the different SMPs and uni-processors used as compute nodes and our aim is to improve the system software used in such clusters. Characterizing SMP Memory Systems Performance As a first step we developed a novel memory system microbenchmark that is independent from naming and coherence issues (i.e. CC SMP, CC NUMA, NON-CC NUMA or Msg Passing) and is therefore capable to characterize the memory system performance for both local and remote memory, regardless of the underlying architecture. Unlike the simplisti