[c-nsp] C2960G and output drops

Gert Doering gert at greenie.muc.de
Thu Oct 2 03:50:34 EDT 2008


Hi,

On Wed, Oct 01, 2008 at 10:46:18PM +0200, Hans Verkerk wrote:
> I faced similar problems on 3750 platform and also resolved it with
> etherchannels (under time pressure). I took some time reading QoS
> details of 3750 platform, but did not take time to test in real world.

OK, I've done quite some testing, tuning "to the extreme"...

Queueset: 1
Queue     :       1       2       3       4
----------------------------------------------
buffers   :       1       5      93       1
threshold1:     100     200    1600     100
threshold2:     100     200    2400     100
reserved  :      50      50      90      50
maximum   :     400     400    3200     400

(and all traffic is mapped to queue 3).

> *** Maybe *** your problems are caused due to lack of Tx queue buffers.
> By default all buffers are equally shared over all four Tx queues:

It definitely looks like it - by tuning the buffers, I can change the
drop rate between "too high" and "catastrophic".

Without mls qos, it seems to just give the port *all* the buffers, with
no fancy "take away a few buffers for this and that" schemes going on - 
so the drop rate is actually lowest if I turn *off* mls qos...


As a next test, I'll try to move around machines between port ASICs
- maybe things improve if ingress and egress ports are (not) on the same
ASIC.  We'll see.

gert
-- 
USENET is *not* the non-clickable part of WWW!
                                                           //www.muc.de/~gert/
Gert Doering - Munich, Germany                             gert at greenie.muc.de
fax: +49-89-35655025                        gert at net.informatik.tu-muenchen.de


More information about the cisco-nsp mailing list