On 09/22/2011 03:00 PM, Jeffrey Altman wrote:
On 9/22/2011 5:08 AM, Anders Magnusson wrote:
What I meant here is that the gateway is a Centos 5 machine running
Samba and OpenAFS 1.4.10 and store data in AFS. So, writing to the
same place in AFS via this machine is significantly faster than using
the IFS client. This is just to show that the performance problem isn't
there if using the Linux client.
If the behavior you are seeing is not a regression from the OpenAFS
Windows 1.6.0 client on your systems, then the performance limitation is
simply a design limitation of the current Windows cache manager
implementation. The cache manager currently requires that all StoreData
operations be serialized which limits the number of simultaneous
StoreData RPCs on a file to 1. Small chunksizes will result in poor
performance. I use
ChunkSize 23 (decimal)
which equates to 8MB per RPC.
I have tested with the 1.6.0 client, and get about double speed compared
to the IFS client.
I also tried to increase the chunksize as suggested, but no difference
in performance.
So, this is still a mystery for me :-/
-- Ragge
_______________________________________________
OpenAFS-info mailing list
OpenAFS-info@openafs.org
https://lists.openafs.org/mailman/listinfo/openafs-info
