Dale,
Please review
http://www.cisco.com/en/US/products/hw/routers/ps133/products_tech_note09186a0080094791.shtml#topic4
and for buffer tunning ..
http://www.cisco.com/en/US/products/hw/routers/ps133/products_tech_note09186a00800a7b80.shtml
Working in a high impact financial services environment for many years, we
see output/input queue drops all the time on various platforms, hardware,
linecards etc. Especially when dealing with bursting market data traffic
egressing an interface towards a firewall or a host.
Now its quite possible for you to have 0 buffers available as the reason for
the output queue drop, but in Satit's case, the interface recorded no such
buffer misses. And quite frankly, with increased buffers sizes on platforms
across the board, you'll rarely see in today's world (except for the low end
routers) buffers misses as the cause for output queue drops.
Finally, while Juniper states on the docs that the ISG1000 is capable of
handling 1Gps throughput, lets just say (no knock on them), we've recorded
drops at much lower sustained transmission rates.
On Sun, Oct 24, 2010 at 5:13 PM, Dale Shaw <dale.shaw_at_gmail.com> wrote:
> Hi,
>
> On Mon, Oct 25, 2010 at 6:52 AM, Abdul <rslab007_at_gmail.com> wrote:
> > Output que drops are a function of the recipient host (in your case, the
> firewall) unable to keep up with the data that is being sent by the switch.
> I am sure if you have the switchport enabled to send and receive Ethernet
> pause frames, you would see the a ton of input frames coming from the
> firewall.
> >
> > This problem isn't with your switch unable to keep up. It's with your
> firewall. Sure you could do things like policing your outbound traffic, but
> your probably better taking a more higher scope view as to what is the best
> QOS policies would be best suited and where (i.e where to mark your traffic)
> in your environment. Then executing your policies.
>
> I have to disagree. Output drops are caused when the switch runs out
> of buffers. 2960 is notoriously bad at coping with bursty traffic. I
> doubt it's got much at all to do with the firewall.
>
> Also, I recently read a post suggesting that enabling QoS on a 2960 is
> very likely to make it problem worse because it essentially causes the
> small amount of buffers to be carved up between 4 tx queues (I have no
> hands-on experience with c2960 so this is not authoritative).
>
> Cheers,
> Dale
Blogs and organic groups at http://www.ccie.net
Received on Sun Oct 24 2010 - 18:13:58 ART
This archive was generated by hypermail 2.2.0 : Mon Nov 01 2010 - 06:42:06 ART