[ Friday, February 25, 2000 ] Gregory Leblanc wrote:
> Does anybody have a nice script to run tiobench through it's paces, and
> collect results? I'm going to try that SPARC patch tonight, and run RAID5
> through it's paces. I suppose I'll need to find a nice way to keep the
> processors busy, maybe compiling something into /dev/null...
the funnyscripts/ directory (hopefully soon to renamed just scripts/)
has some stuff to get the results into gnuplot. tiobench output by
default is fairly good, and it can go through all combinations of
specified parameters.
> Isn't load avg just the number of jobs that are waiting? My mail servers
> have a load avg of about 5, but they're not at all CPU bound, it's all about
> I/O. Ahh, how I wish I had linux on that DS20 in the back room...
load avg is avg # of jobs in the run queue, which processes blocked on
I/O shouldn't be. load avg > # of processors should mean there's always
enough processes ready to keep the CPU's busy (at least on average),
so while individual processes may not be CPU-bound, I don't know of
a scenario where the machine could get to that state and not have the
processors pegged.
> Well, our most heavily connected machines here have a single 100MBit
> connection. Anybody have some gigabit (copper of fiber) equipment they want
> to loan me for some testing? :)
I've got a quad xeon with dual gnic-ii hamachi's running over crossovers
to some dual xeon's, but it'll be a couple of months (after I finish
my semester's work on cluster-based virtual filesystems) before I can
donate some time to s/w raid analysis... hmmm
James