[PVFS-users] PVFS on a 14 CPUs cluster

Robert Ross rross@mcs.anl.gov
Tue, 3 Oct 2000 14:35:17 -0500 (CDT)


I've run tests using more than 60 compute processes, so that isn't the
problem in and of itself.

Are the I/O servers using DHCP as well?  The manager node?  It's possible
that one of them got a bad address.

What sorts of errors are you getting?  Which method are you using to
interface with the file system?  Did you recently change the kernel or the
libc installed on your clients?


Rob Ross, Mathematics and Computer Science Division, Argonne National Lab

On Tue, 3 Oct 2000, Ibrahim Haddad (LMC) wrote:

> I have been testing with PVFS on a 7 CPUs cluster for a while
> and it went ok. Now I added 7 more processors to have a total
> of 14 processors. Things are not ok anymore.
> The CPUs are diskless and they use NFS as a file space area. 
> They get their IP via DHCP and download a ramdisk that is
> similar on all of them. When I had 7 processors only, I didnt
> get any problem. Now, with 14 processors, I am not able to
> access files within the PVFS file system from all CPUs. I
> am not able to copy from all CPUs...
> I have 1 IO node. I had 3 but I minimized it to one while trying
> to figure out why it is behaving like that. And I have 1 management
> node. 
> Did anyone try out PVFS with a high number of clients (>=14)??