[c-nsp] 2960 -> 4948 - no more drops :)

[Alex Pressé]

Not sure if guerrilla marketer trying to get readers to google this
fantastic switch...


On Sat, Feb 16, 2013 at 5:14 PM, CiscoNSP_list CiscoNSP_list <
cisconsp_list at hotmail.com> wrote:

>
>
> Hi Rob (Sorry for not replying inline, but hotmail screws the formatting)
>
> We did try tuning qos buffers (It did improve the drops, but they were
> still significant), and we also tried disabling mls qos (Still saw
> significant drops)....Im really interested to know why there is such a
> difference between the two platforms....i.e. is it buffers/how they are
> allocated, architectural differences or combination of both?
>
> Cheers.
>
>
>
> > Date: Sun, 17 Feb 2013 00:27:31 +0100
> > Subject: Re: [c-nsp] 2960 -> 4948 - no more drops :)
> > From: robhass at gmail.com
> > To: cisconsp_list at hotmail.com
> > CC: cisco-nsp at puck.nether.net
> >
> > > We recently upgraded a 2960G(Only doing L2) that was hitting
> ~500Mb/sec on one port, and we were seeing 40,000+ output drops (5Min) -
> Since the swap to the 4948, we see zero output drops. Is the difference in
> performance purely buffer size?  I *think* the 2960 has 1.9Mb (Per ASIC)
> and the 4948 has 16Mb (total?)?
> >
> > It can also be default srr-queue configuration if mls qos was enabled.
> >
> > Try connect host again to 2960G but configure 'srr-queue bandwidth
> > shape 0 0 0 0' on all ports before.
> >
> > You can also :
> > - assign all traffic to one particular queue which increase amount of
> > buffers (eg. all dscp's to queue2)
> > - reconfigure (increase) thresholds for queue2
> >
> > 1900000/24 ports = (79166 / 4 queues) * 8bits = 160Kbit per queue
> >
> > No so much. IMHO Cat4948 has more dynamic buffering instead of static
> > allocation per port / per queue.
> >
> > Rob
>
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
>



-- 
Alex Presse
"How much net work could a network work if a network could net work?"