[perfmon2] papi on compute node linux
Status: Beta
Brought to you by:
seranian
From: Mark W. K. <kr...@cs...> - 2008-03-10 16:21:58
|
I've been testing the recent PAPI commits, and they seem to solve the dropping interrupts problem with perfmon on standard Linux. You may recall, we've seen a similar problem on Compute Node Linux at ORNL and NERSC. The symptoms are the same: at some point, something happens, you just stop getting interrupts and they never resume. On CNL, this problem happens without threads. This morning (2008-03-10), I tried the latest PAPI from CVS, compiled for CNL on jaguarcnl with the options: ./configure --with-OS=CNL --with-perfmon=22 --prefix=... It compiles, links and runs, but still drops interrupts. Take a look on jaguarcnl at: jaguarcnl: ~krentel/papi-cnl, ~krentel/papi-2008.03.10 There are similar directories on franklin at NERSC. Stephane, would you be willing to take a look at CNL? Phil, I'm guessing that the PAPI change to use SIGRTMIN instead of SIGIO would also apply to CNL, since it's also using perfmon, right? Any other configure options to try? Thanks, --Mark |