Murali Vilayannur wrote:
Nice investigation Phil! Results look great. Isnt this approach similar to what Julian was doing with the threaded I/O work in the kunkel branch? (with the exception of the O_DIRECT stuff?)
I'm not sure- I haven't actually been following those commits that closely yet. This is something we just cobbled together a while back as an experiment but weren't really sure if it was useful at the time (it failed to fix a different problem that we were looking at which later ended up being a kernel bug). We only yesterday ran tests that showed it being helpful for a real world scenario.
It also looks like the trove backend could benefit from the readx, writex posix extension patches to the kernel and ext3, although that would be a lot of work I guess..
Yeah, probably so. I doubt we will be able to work with it in our environment until some of the major vendors ship kernels with readx/writex available...
-Phil _______________________________________________ Pvfs2-developers mailing list [email protected] http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers
