[cifs-protocol] Re: [linux-cifs-client] Linux CIFS performance

Steve French smfrench at austin.rr.com
Wed Mar 7 23:20:47 GMT 2007


> -----Original Message-----
> From: cifs-protocol-bounces+lam=synplicity.com at cifs.org
> [mailto:cifs-protocol-bounces+lam=synplicity.com at cifs.org] On Behalf Of Dave
> Kleikamp
> Sent: Wednesday, March 07, 2007 11:12 AM

> 
> Did you try dd?  Of the two things my patch does, the part I was trying
> to fix was to address writes that were smaller than the page size and
> beyond the end of the file.  I expect that to help dd with the small
> block sizes.  The other fix for the bug I stumbled upon probably doesn't
> affect dd performance.
> 
> I'll try to run a few tests myself.
> 
> Shaggy

My quick results do confirm that this patch also helps cifs with dd
(although not enough to beat NFS  - still need additional improvements),
SLES10 server, and client kernel going from about 2.6.21-rc1 vs. about
rc3  (which includes Shaggy's perf fix)

(current kernel 2.6.20rc3 ie with patch)
smf-t60p:/cifs-with-patch # time dd if=/dev/zero of=/cifs/.test bs=1024
count=250000
250000+0 records in
250000+0 records out
256000000 bytes (256 MB) copied, 34.9132 s, 7.3 MB/s

(without patch)
smf-t60p:/cifs # time dd if=/dev/zero of=/cifs/.test bs=1024
count=250000
250000+0 records in
250000+0 records out
256000000 bytes (256 MB) copied, 77.5971 s, 3.3 MB/s


dd performance over GigE with dd blocksize of 1K - increased from about
3.3MB/sec to over 7MB/sec (more than doubled). I repeated it multiple
times each way to confirm.   This server was a laptop (slow disk).
Shaggy saw much bigger improvements in dd performance on his test
systems.  

As dd block size increases I would not expect the patch to make as much
difference, but over slower networks the improvement might be much more
dramatic.

The reason for the difference was obvious: before the patch the writes
were 1K in size over the network (thus 250,000 of them) - with the patch
the writes were much larger (and thus fewer than 5000 of them - so
network latency was much less a factor and the server was a lot more
efficient).

Yeah Shaggy!



More information about the linux-cifs-client mailing list