[cifs-protocol] Re: CIFS/NFS performance of netapp filer vs local dir

Matt Seitz seitz at neopathnetworks.com
Thu Mar 1 16:25:21 GMT 2007


Joseph Kuan wrote:
> The main reason of poor network throughput because when you open a file, 
> is not a simple open, read, and close. 

Actually, that is all the CIFS protocol requires.

> It sends lots of TRANS2 packets 
> and the same file open and close several times before reading the whole 
> files, lot of other weird network operations such as NTFS data stream 
> checks.

Those are additional operations being requested by an application.  There's 
nothing in the CIFS protocol that requires that.

> I am in the middle of the process documenting on this issue. Will let 
> you know more details in near future (if you are interested).

I would be interested in seeing your results.  I suspect what you are seeing is 
either a) poor application design, b) poor CIFS client design, or c) use of 
features that are simply not supported by NFS.

Now, in practical terms, that may mean a particular client sees better 
performance with CIFS than NFS.  But I strongly suspect that is because of the 
CIFS client implementation, not any inherent difference in the protocol.


More information about the cifs-protocol mailing list