[Bioclusters] BLAST Memory Benchmarks revisited

[Juan Carlos Perin]

I am once again playing with our Xserve cluster and BLAST searching large
databases.  My benchmarks focus primarily on the NT database.  I've run
comparisons against similar machines, with slight variations in hardware
configurations.  My questions are specifically about the G5 architecture.

First, I noticed that after upgrading a single node to 4 gb of RAM,
essentially 2gb per processor, the speed of blast searches increased about
10 times, in comparison to identical runs with only 2gb (1gb/processor).
Has anyone had any experience with more than 4gb total?  I don't notice
virtual memory being accessed after upgrading the memory, so I am assuming
that this is the minimum amount of RAM required to see this kind of a
performance increase.  We are curious to see if 'even more' memory would
continue to improve speed, or if 2gb is 'sufficient'.  My assumption is the
latter, but just wanted to see if anyone had any hard evidence.  We are
limited to this amount because we only have 512 chips, and the Xserves only
have 8 slots...

Second, to accommodate this memory restriction, and perhaps test this on our
own, I was considering removing a CPU as to force all the memory slots to be
allocated to a single CPU.  I am wondering if this would actually work? Or
if the architecture is actually segregated and each 4 slots is for an
individual CPU?  


Juan Perin
Bioinformatics Core
Children's Hospital of Philadelphia