[cvsnt] Strange CVS performance fluctuation

Arthur Barrett arthur.barrett at march-hare.com
Wed Jun 21 00:35:59 BST 2006


Community technical support mailing list was retired 2010 and replaced with a professional technical support team. For assistance please contact: Pre-sales Technical support via email to sales@march-hare.com.


Trevor,

Oops - hit send before I'd finished...

Thanks for publishing the stats - they are quite interesting.

Can you answer a few more questions about your test?:
* was the local RAID tests performed with hardware or software RAID?
* what was the architecture of the local disk (IDE, SIDE, SCSI, SCSI
Ultra 320, SCSI Ultra 160 etc)?
* what was the architecture of the local IO (PCI, PCIX, ATBUS etc) and
speed (if known)?
* what was the file system on the array (NTFS, FAT, other)?
* was a battery backed write cache installed the local RAID controller?
* was a battery backed write cache installed the SAN ?
* what is the SAN interconnect (eg: iSCSI, Fibre)?
* what were the SAN disks (SIDE, SCSI, SCSI Ultra 320, SCSI Ultra 160
etc)?

Also a little more information about the test itself?:
* I assume the checkout was to another windows client, which OS was it?
* what was the interconnect between the client and the server (eg:
100Mb, Gigabit)?
* number of processors / processor speed / bus speed on the server?
* number of processors / processor speed / bus speed on the client?
* what protocol was used (SSPI, PSERVER, SSH)

I am thinking a better test would have been to use the "remote protocol"
such as SSPI, but on the same server.  I think the bottom limit of 16
minutes or so probably was the maximum throughput of your network - so
one of those RAID 5 configs (SAN or Local) may have actually been
"faster" but it cannot be determined from those stats.  Also a command
like "rlog" may have more effectively tested the "server side" disk IO
(read) performance more than something like "checkout".

> The CVS Server is 2.5.02.2115 and the client was 2.5.03.22650

>From memory recent builds of CVSNT Server 2.5.03 do address a couple of
performance issues, but not specifically disk performance.  In
particular in 2.5.03 server "atomic checkout" is now disabled by default
for optimal performance and stability, so it would be interesting to try
the same tests with the newer server when it's declared stable.

Thanks,


Arthur



More information about the cvsnt mailing list
Download the latest CVSNT, TortosieCVS, WinCVS etc. for Windows 8 etc.
@CVSNT on Twitter   CVSNT on Facebook