Poor PE4600 NFS performance

James Bourne jbourne at mtroyal.ab.ca
Mon Aug 5 22:38:00 CDT 2002


On 5 Aug 2002, Rick Gaudette wrote:

> Hi Jason,
> 
> Thanks for he tips, I have some questions regarding them.
> > 
> > not sure what advice you've already got, here are some tips on increasing
> > your nfs performance - some or all might be applicable.
> > 
> > o is this an intel gigabit card ? consider using the latest drivers (4.3.2)
> It is the onboard broadcom chip.
> 

Quick question, how close are you to your switch (cable length, end to end)?

IIRC, copper gigabit is only good for 100 meters maximum...  Likely you're
OK there but...

Also, what is the output from ifconfig eth? for the gigabit device?
Specifically the lines:
          RX packets:6824497 errors:0 dropped:0 overruns:0 frame:0
          TX packets:2321723 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:100 

Regards
James Bourne

> > 
> > o increase the number of nfsds available (128 is the normal maximum, which
> >   will increase soon i think)
> Would this still apply if there are typically only one or two users
> reading at any given time?  Our typical use model is that one or two
> users might be trying to read a 1-2 GB data file at any given time. 
> That is the type of use model we wish to get the best performance for.
> 
> > 
> > o if you have end to end gigabit between your server and clients, you could
> >   consider increasing the MTU to 9000
> 
> Yes end-to-end gigabit for all of the clients.  I have thought of it but
> have two concerns.  Some of our clients have the NS83820 based card and
> I have heard that they don't support jumbo frames well.  Also, if jumbo
> frames are enabled can the machines access the rest of the net?
> 
> > 
> > o are your clients using 1K, 4K, 8K or 32K mount sizes ?  bigger sometimes is
> >   worse!
> 
> Anyone have any recommendations here?  We have tried a bunch but the
> default seems to work best.  The man pages say the default is 8K but
> when we specify 8K we get degraded performance.  The defaults seem to
> perform the same as 32K.
> 
> >
> > o are you clients using nfs over UDP or TCP ?
> 
> UDP.  I would like to try to use TCP but as far as I can tell the RH
> kernels don't support it.  Anyone know how to enable TCP based NFS on RH
> kernels.
> 
> A bit more info that I have noticed.  nfsstat reports that the number of
> retries is pretty significant compared the number of calls.  The clients
> all say that there is 1 retrans for every 3 calls to the PE 4600.  This
> doesn't sound good to me but I am not sure how to fix it.
> 
> Thanks
> Rick
> 
> 
> > 
> > regards,
> > 
> > -jason
> > 
> > _______________________________________________
> > Linux-PowerEdge mailing list
> > Linux-PowerEdge at dell.com
> > http://lists.us.dell.com/mailman/listinfo/linux-poweredge
> > Please read the FAQ at http://lists.us.dell.com/faq or search the list archives at http://lists.us.dell.com/htdig/
> 
> 
> _______________________________________________
> Linux-PowerEdge mailing list
> Linux-PowerEdge at dell.com
> http://lists.us.dell.com/mailman/listinfo/linux-poweredge
> Please read the FAQ at http://lists.us.dell.com/faq or search the list archives at http://lists.us.dell.com/htdig/
> 

-- 
James Bourne, Supervisor Data Centre Operations
Mount Royal College, Calgary, AB, CA
www.mtroyal.ab.ca

******************************************************************************
This communication is intended for the use of the recipient to which it is
addressed, and may contain confidential, personal, and or privileged
information. Please contact the sender immediately if you are not the
intended recipient of this communication, and do not copy, distribute, or
take action relying on it. Any communication received in error, or
subsequent reply, should be deleted or destroyed.
******************************************************************************


"There are only 10 types of people in this world: those who
understand binary and those who don't."





More information about the Linux-PowerEdge mailing list