Hello,
I am measuring the performance of OrangeFS and I looking for
documentation on the writing part.
When writing more than 100 MB on one OrangeFS server (with ROMIO), the
measured bandwidth is dropping. This corresponds with the behavior for
any size when TroveSyncData is enabled. It seems that there is a buffer
for storing write requests and when it is full, data is actually flushed
to the disk. I assume that this is a normal behavior but I did not find
info regarding this (how to select this buffer size for example).
The problem with this situation is that the resulting bandwidth is poor
and unstable: between 5% and 50% of the peak bandwidth despite being
close to the maximum when reading data (or writing less than 100 MB with
TroveSyncData disabled). I got this with ext3.
I am interested to understand those performance, but I can provide
details if anybody suspect a misconfiguration.
Thanks in advance,
Louis
_______________________________________________
Pvfs2-users mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
