[c-nsp] Possible memory corruption w/ 12.2(18)S11

Rodney Dunn rodunn at cisco.com
Wed Mar 22 08:14:01 EST 2006


> 
> I've always been inclined, (based on past experiences w/ CCO) to run the 
> SERVICE PROVIDER/VIP loads of IOS. I was told that this is tuned for 
> Providers in several ways and makes the best choice for running high 
> availability BGP routers.

Not really. The feature sets are twofold. a) It's a market differentiation
for pricing b) It doesn't force the entire feature set and code bloat
on customers that don't want it. Both are valid things that are needed.
The feature set distribution isn't perfect but they do try and get
it right. That's why the new image reformation did new feature sets.

As for service provider. The idea way back on that was to not provide
a bunch of features in an image that wasn't needed by the big 
service providers (NAT, some security stuff, etc..). Now a lot
of those features over the years have merged in the market place
and become required in both customer spaces so it's no as easy
to keep them separate anymore.

That being said, 12.0S vs. the other trains (12.2,12.3,12.4) is
a different story. The protocol (mainly BGP) implementations in
that train are *very* targeted towards high density BGP designs
because that is the code that the big service providers run.
A lot of it is history and the fact that the GSR runs there.

Some of the BGP features developed there do make it back in to
trains like 12.2-12.4 as double commits. It just depends on the
need of the feature in platforms that only run in those trains.
ie: we don't need a 1k BGP peer convergence improvement on a train
where that would never exist...someone might try it but we don't
recommend it so we don't encourage it.

All that being said the general direction is if you have a large
network running in the core that is heavily BGP then 12.0S for
scalability is probably the best place to be. As the 76xx has
moved in to the SP edge and even core space in some setups the
features from 12.0S are showing up in those trains (SXE,SXF, and
forward more will come) that run on those platforms.

My general "What code is best?" answer to customers today is this:

a) Large scale BGP design or you want HA on a 75xx (12.0S). (26S, 27S..)
   There may be some features you can't have though (like IRB, etc..)
   in 12.0S. That's where the 75xx is too flexible becuase all the
   features like that are not supported in one train. They never will be.
   We recommend 72xx,7301's for IRB type stuff.
b) If you are using it just as a leased line aggregation and don't need
   heavy BGP stuff (I'm talking hundreds or peers and full feeds) the
   other trains such as 12.2, 12.3, and 12.4 should do just fine). Even
   with a couple of full BGP feeds. It's when customers want to run fancy
   broadband aggregation stuff and other features the box was never designed
   to do that they get in trouble. The only reason the features were there
   was because they were configurable. We try to prevent that when we can
   becuase it just leads to customer dissatisfaction when it doesn't work
   because we never tested it.

The 75xx is absolutely the most complicated box to pick code for
because it's been around soooo long and is used in so many places.

I actually have a meeting with the business unit to try and come up
with a forward looking IOS release recommendation for customers on
this platform because my view is customers need to understand where
the platform and code is headed. Today I don't think that's done.

I'll let you know the results.

Sorry it was such a long response. I just got my morning dose of coffee.

Rodney

 
> Would you agree with that?
> 
> And secondly, what are the major differences between the follwing release 
> trains? If there is a document that explains this, I'd be happy to read 
> it.
> 
> IP 
> IP PLUS IPSEC 3DES 
> ENTERPRISE IPSEC 3DES 
> ENTERPRISE 
> SERVICE PROVIDER 
> SERVICE PROVIDER/VIP 
> 
> 
> -- 
>     Vice President of N2Net, a New Age Consulting Service, Inc. Company
>          http://www.n2net.net Where everything clicks into place!
>                              KP-216-121-ST


More information about the cisco-nsp mailing list