Did you download 1.3.2, and did that fix the other problems?
You could consider using the --enable-epoll option as well; this will
make for a more efficient server in cases with lots of clients like this.
If your files are relatively small, or if they are all used only by one
process at a time, in the StorageHints section of the file system
configuration file you should add "DefaultNumDFiles 1" like this:
...
<Filesystem>
...
<StorageHints>
TroveSyncMeta yes
...
DefaultNumDFiles 1
</StorageHints>
</Filesystem>
...
Those things should all help.
Rob
Uday Shingwekar wrote:
Hi,
The system has heavy load (the pvfs2-server cpu% being over
50-90 range) with over 800+ clients. The amount of data is over 1.7TB in
the file system. The total number of files/symlinks/directories are over
8000. The number of symlinks is over 4000+. The application typically
accesses the files through the symlinks. I am not sure about the block
size that the application uses, but I would assume that it is small i.e.
about 4k-8k range.
Thanks,
uday
-----Original Message-----
From: Rob Ross [mailto:[EMAIL PROTECTED]
Sent: Tuesday, December 13, 2005 8:13 PM
To: Uday Shingwekar
Cc: [email protected]
Subject: Re: [PVFS2-users] Large "stat" times for files on the pvfs2
cluster, AttrCache settings
Hi,
What else is going on on the system, what is in the directory, etc?
It's fairly unusual for things to take this long.
Rob
Uday Shingwekar wrote:
Hi,
We are seeing that the "stat" call for the files on the pvfs2
cluster
takes over 2 seconds. Has anyone seen timings like these? Is there an
optimization setting that can improve this performance?
What are the suggested optimizations referred to in the FAQ
specifically related to AttrCache settings on the server?
Also, what are the settings on the client if the application can
deal
with out-of-sync caches?
thanks,
uday
_______________________________________________
PVFS2-users mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
