Hello,
With PVFS, can I expect to get speed-up in the 2 following cases:
1. Copy data (with scp or rsync) from local storage (my laptop) to the
PVFS storage.
If yes, do I have to do some specific tuning ? (ssh config file ?
others ?)
2. Run a code that uses MPI-IO (or HDF5) to read or write (big) files.
If yes, do I have to modify my code and / or re-compile MPI (or
HDF5) with specific options to account for PVFS ?
I am completely new to file systems related stuffs. I came to PVFS by
googling "open source parallel file system" as I try to understand why
we observe speed-down behaviors when using our CentOS+GPFS cluster (I
suspect the problem is related to the file system).
At the very first googling, my understanding was that, as soon as PVFS
is installed and the PVFS server is running, I should get speed-up when
copying data and also when running my code (my code uses MPI-IO or HDF5
+ MPI-IO and HDF5 should be speeded up by the PVFS server) : my
understanding was that in some ways the PVFS server could speed-up
things (copy data, running) without compelling me to change anything in
the usual copy / run process. When I went on reading the doc, I read
about libpvfs : I am not sure to understand why / how one needs to use
it. Do I need to re-compile MPI (HDF5) taking account for libpvfs ? Do I
have to recompile my code taking account for libpvfs ? I am not sure to
understand how things have to be done : this is why I post this mail to
get some help / explanations.
Thanks,
Franck
PS : I run Ubuntu 14.04 on my laptop. For now I only would like to
install PVFS on my laptop to see if I can get some speed-up. If I
succeed to use PVFS on my laptop, I may try to use it on the cluster to
compare with gpfs (this is why I googled "open source parallel file
system" !).
_______________________________________________
Pvfs2-users mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
