Dan, While I'm looking through the data you provided, can you create a UFS file system on the x4500 and try the script against it? This will help identify whether or not ZFS is a factor.
Thanks, Alan -- On 10/03/08 07:42, Daniel Liebster wrote: > Here are the relevant files, the python script, the cifs-gendiag output, and > a screenshot of the traffic flow during the script's execution. The part of > the traffic flow that exibits a spikey pattern with a relatively small area > under the curve is the script working against a Windows server, and then the > HW filer. The two parts that follow with very high bandwith utilization occur > when the script works agaist the 1Gb interface, and then the 10Gb interface > of the x4500. The bandwidth against the x4500 is very high, and the script > takes twice as long against the CIFS-server compared with others. > > I can't install Windows server on the x4500, as it needs to export NFS > volumes to out HPCC as well as CIFS to scientists desktops. > based NFS has been tried by others in my group, but no solution was found > that could adequately handle the NFS workload that our genomics cluster can > generate. > > The windows fileserver used in this test is an 2 year old Aberdeen SATA > server(Areca raid controllers).. > > > Kernel version: Microsoft Windows Server 2003 R2, Multiprocessor > Free > Product type: Standard Edition > Product version: 5.2 > Service pack: 2 > Kernel build number: 3790 > Registered organization: Cold Spring Harbor Laboratory > > Processors: 4 > Processor speed: 3.2 GHz > Processor type: Intel(R) Xeon(TM) CPU > > In terms of basic file copy IO, the Aberdeen is about 30% slower than the > x4500 in robocopy tests I did. The traffic patter that this python script > generates is random access to data in a 60MB file. Apparently this pattern > runs up against something in CIFS-server that makes it perform at half the > speed that a HW or Windows based server can. > > Please let me know if there is any type of information I can gather to help > get to the root cause. > > Dan > -- > This message posted from opensolaris.org > > > ------------------------------------------------------------------------ > > _______________________________________________ > cifs-discuss mailing list > [email protected] > http://mail.opensolaris.org/mailman/listinfo/cifs-discuss _______________________________________________ cifs-discuss mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/cifs-discuss
