[nsp] 12000 OC12 ATM Card

[Jared Mauch]

On Sun, Mar 09, 2003 at 04:20:12PM -0800, warner at cats.UCSC.EDU wrote:
> >Daniel Puka said:
> >
> >    (Customer 1,2,3)-> (STM1-ATM SWITCH) <-> (OC12 CARD - CISCO 12000)
> >
> >     The cisco documentation says that when one vc get congested it can
> >use resources from other vc in the same card because it has only one
> >single queue for all vcs.
> >
> >     One week ago I started to experience very high packet delay during
> >the peak usage hours(about 150ms where it was usually 1ms). I am not
> >experiencing any packet, cell or crc errors. Is there any change that
> >one vc congested from other customer can be the cause of delayed
> >packets? My maximum usage is about 90Mbps.
> 
> Well, there's always a chance...  But it seems unlikely.  You
> should find the delay when you ping your service provider's OC-12
> interface that faces you.  If this is gives 150 mS response times
> when you're seeing general high packet delay, you should look 
> further at the ATM edge stuff.  My guess is that you'll find that the
> problem is not in your ATM distribution.  UC Santa Cruz has a very similar
> edge setup -- except that in your case the ATM is _not_ oversubscribed.
> Every time we have seen this kind of net slowdown (without packet loss)
> it has been elsewhere in the service provider's network.

	Assuming you are talking about the Engine0 OC12 ATM card,
they have quite poor performance if you enable any features.  These
include things such as:

	1) Unicast RPF (either strict or the loose check)
	2) Interface ACL
	3) CAR of any sort, including CAR against acl.  (I believe
that Cisco calls it PIRC)

	I'm sure there are other features that may impact
things, including the "DOS Tracker".

	- Jared

-- 
Jared Mauch  | pgp key available via finger from jared at puck.nether.net
clue++;      | http://puck.nether.net/~jared/  My statements are only mine.