On Oct 4, 2007, at 4:42 PM, Sam Lang wrote:
On Oct 4, 2007, at 3:45 PM, Pete Wyckoff wrote:
[EMAIL PROTECTED] wrote on Thu, 04 Oct 2007 15:21 -0500:
I'm in the process of getting a trace and dump from the same server
during the same runs.
Good, will be interesting too see tcpdump + strace.
Pete,
Here's a trace and dump from the same set of runs.
http://www.mcs.anl.gov/~slang/epoll-trace.out [370MB]
http://www.mcs.anl.gov/~slang/server-packets.dump [180MB]
Again, the trace isn't from strace, I'm tracing the epoll calls
manually.
Damnit, I forgot to recompile the servers. These results are with
poll. The slowdown still exists for these runs though.
-sam
-sam
Although I
worry we might not learn anything new.
In the meantime, I attached two "zoomed" plots of the last one I
sent, with ranges set to 5000-5200 secs and 5000-6000 secs,
respectively. I thought they were interesting, both for the
behavior
that epoll_wait exhibits during operations over the VFS, and just
for
the patterns they show at a scale of 100s of seconds (it almost
looks
like a seirpinski triangle). Not that it sheds much light on the
problem.
Pretty. Maybe there's a category for Most Attractive Unexplained
Distributed System Art in the state fair this year.
Would the pending request ID stuff with slog etc be useful in
tracking individual operations?
-- Pete
_______________________________________________
Pvfs2-developers mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers
