[nsp] 12.0(22)S - avoid it

[Leon Chang]

Unfortunately I heard about 12.0(22)S actually is 12.0(22)ST after I have
upgraded our core routers to 12.0(22)S, this was intended to solve the
dCAR on dot1q interfaces bug. A 7576 crashed 14 hours after
upgrade right in the middle of peak time with output stuck and cbus
complex on FE interface. We have reverted back to 12.0(21)S3.

I am very suprised that Cisco has decided to merge 12.0ST train into S
train without telling people exactly what they have done since stability
is supposed to be the top priority for S-train and people upgrade to the
latest version to avoid known bugs and hoping to improve stability.

In your case, 12.0(21)S3 is still the most stable version from my
experience, I would rather run 12.0(21)S3 and move the affected interface
to another FE port on a separate VIP card to work around the dCAR bug.

Regards,

Leon Chang, CCNP, CCDP
Senior Network Engineer
The Internet Group Ltd. (IHUG), Auckland, New Zealand  AS7657
PH:  +64 9 359 2710
---------------------------------------------------------------------

On Sat, 24 Aug 2002, Antoine Versini wrote:

> Hello,
>
> A few days ago, I upgraded five GSR from 12.0(19)S2 to the brand new
> 12.0(22)S (the one that merges the S and ST trains)... What a mistake !
>
> 1/ Some STM-16 links between involved GSR became to flap. Got "Loss of
> signal" one side and "Loss of frame" other side, got a _huge_ amount of
> "Input errors" even when doing a "loopback internal"... The links flaped
> more when they where taken out of production, interresting.
>
> 2/ Bogus BGP advertisements appeared. A "show ip bgp neighbor X
> advertised" was correct, but the peer actually didn't received any of
> the prefixes. "clear soft out" or "clear out" didn't solved the issue, I
> had to reset the session. This happen on different routers, on different
> IX...
>
> 3/ A "neighbor X weight Z" (where X can be an actual BGP peer or a peer
> group) is accepted by the router but not shown in a "show run" nor
> applied. But an inbound route-map with a "set weight" was still working.
>
> I also noticed some strange latencies in ssh or telnet (you known, when
> the charaters you type are echoed slowly). The averaged CPU usage was 8%
> before, 11% after the upgrade. So is the memory usage, 20MB more than
> with the previous version...
>
> Well, all that I can say is that this version has some issues (and, yes, I
> was foolish to put in production a 12.0S version with no digit behind the
> 'S' ;-) ). Now reverting to 12.0(21)ST3. Too bad, the new SSO (successor
> of RPR+) was very promising...
>
> If someone had any issue with 12.0(21)ST3, please tell me!
>
> --
> Antoine Versini
>
> _______________________________________________
> cisco-nsp mailing list  real_name)s@puck.nether.net
> http://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
>