[c-nsp] SXH3 ghost bugs - more details

[Gert Doering]

Hi,

On Mon, Sep 15, 2008 at 09:11:14PM +0100, Peter Taphouse wrote:
> Just to second (or third?) this bug.  We've got four 7600s on SXH3 which
>  are afflicted by this - they were upgraded from 2a on tac's advise (to
> avoid netflow bug related spontaneous reloads) - and we don't use
> dampening.  It doesn't seem to matter if the prefixes that get withdrawn
> are i or ebgp, they get still "ghosted" to other ibgp peers.  

For iBGP->iBGP ghosts, our current setup is not suitable (read: no
route-reflector setup, so iBGP->iBGP announcements would not take place),
thus I have no evidence on whether this would also trigger the bug for
us.  But I think it's quite likely indeed, given that this seems to 
happen on sending *out* the withdraw...

> I've got a case open with tac, 

Would you mind sharing the case number with me?  I could forward this to
our TAC engineer so they know "this is not just us".

Do you have a bug ID?

> but it's causing us enough grief that I'm moving back to SXF until 
> things calm down.  

*grumble* - I would love to do that, given that we're quite happy with
SXF since about two years now.  But we were unlucky/stupid enough to
get Sup720-10Gs for these new boxes, and they only run SHX...

> Would love the new netflow stuff in SXH if it gets stable enough...

We're quite happy with the SHX3 netflow.  No crashes (yet, knock wood)
and the load on everything is indeed much lower.

There are some other funnies in SXH3, but these are just annoyances
and not service impacting (we're not using scp).

gert

-- 
Gert Doering
Mobile communications ... right now writing from * Munich Airport *