[c-nsp] VIP4-80 throughput

[Kristofer Sigurdsson]

Hi,

Thanks to everyone who responded.  Comments inline.

2006/1/24, Pete Templin <petelists at templin.org>:
> Kristofer Sigurdsson wrote:
>
> > I was wondering whether or not I could realistically expect full
> > throughput out of putting four STM-1's in two VIP4-80 in the router,
> > a.k.a. two PA-POS-OC3 in each VIP.
> > No other interfaces would be in use, the box would just be routing
> > between the four STM-1's.
>
> Any other features?  MPLS, ACL, uRPF, etc.  This could have a major
> impact on performance.  Do you have other routers upstream that

Minor access lists, MPLS, yes, BGP peering with 1-2 upstreams (full
tables) and a few
iBGP peers with 1 more full table.

> could/would deflect/absorb/melt DOS attacks?

Well, no, this router is an edge router to one of our upstreams (maybe
two in the future).

>
> > The RSP is RSP8, if that helps, we'd be taking two, possibly three,
> > full BGP feeds, IOS 12.0S.  We have 256 MB of memory on the VIP's, if
> > I recall correctly.
>
> Your RSP RAM will get tight at times, but your fundamental limiting
> factor will likely be PPS on the VIPs.  Worst case, put in two more VIPs
> to spread the load.  But I'd guess that you'd be fine (assuming no
> features) with the desired setup.  High-PPS DOS attacks could be
> problematic though, but certainly better than a 7200.

The list seems (well, most at least) to agree that:

1. Three full tables will be tight.  Replace router if/when we add
another upstream or at least take a partial feed...

2. Putting two STM-1's on one VIP4-80 will be tight, especially with
features.  Put the PA's in seperate VIP's.

Thanks again to everybody who responded.

-Kristo