Why SMB2 need asynchronous architecture ?
Jim
yakajob at gmail.com
Fri May 31 19:52:14 MDT 2013
Volker Lendecke <Volker.Lendecke <at> SerNet.DE> writes:
>
> On Thu, May 30, 2013 at 11:57:59PM +0000, Jim wrote:
> >
> > Start from samba-3.6, SMB2 use asynchronous architecture.
> >
> > I don't understand why it benefits the performance of network
application in
> > Windows 7, and what kinds of network application can be benefited ?
> >
> > In my testing in windows7, SMB2 performance is worse than SMB1. (Test
with
> > explorer and robocopy)
>
> In theory, SMB2 has larger packets and less requests,
> leading to higher performance. We have seen reports that
> SMB2 is slower because by default signing is being used in
> SMB2 and not in SMB1. Can you check that?
>
> If that is not the case for you, we need to see why you are
> seeing worse performance with SMB2.
>
> Volker
>
Hi, Volker:
I use the following command to do the test.
robocopy %pc_source%\5GB %syno_target%\ /E /IS /NS /NC /NP /NFL /NDL
I think signature is off because 'Signature' value of SMB2 header is
always 0 during the test.
Test environment:
- CPU: Intel Atom CPU D2700, DDR3 1GB.
- Samba version is 3.6.9
I didn't enable AIO, all the other configuration is by default.
I've test IOMeter on 10G network, the CPU usage of SMB2 is always 100%,
but SMB1 about 8x%. SMB2 seems cost more resource than SMB1.
Thanks,
Jim
More information about the samba-technical
mailing list