Re: [Idr] BGP MIB v2 input

In message <20070326184044.GA11264@scc.mi.org>
Jeffrey Haas writes:
>  
> On Mon, Mar 26, 2007 at 02:21:10PM -0400, Curtis Villamizar wrote:
> > That's not terrible for the NMS.  GET-NEXT on 9.255.255.255/32 and get
> > 10/x or GET-NEXT on 10/7 and walk 10/8, 10/9, etc until you are out of
> > the 10/8 range (in a prefix/mask sense).
>  
> Right, but that means the NMS isn't a very good tool to manage your
> router when looking at routing tables.  People want prefix order, not
> MIB order.

A MIB dump is never a good tool.  A tool that uses the MIB could
reorder such a query.

I've already agreed with you that radix trie order would be better.

> > The ordering in effect requires two trees in the router, one for the
> > normal radix trie acxess and one for the MIB.  At least that is what
> > one implementation does.  Fixing the ordering would eliminate that.
>  
> Every router that implements the MIB would have to do something like
> this.  Sane ones (presumably IMO) all just keep the routes on two trees.
> Less sane ones that are less sensitive to scaling may have copies of
> routing information.

Extra copies?  Might be worth taking this off list.  Could have humor
value.

btw- Really bad first implementations might try to scan the routing
table for each GET or GET-NEXT.  There is existance proof of this.

> > IMO Its useful for enterprise but currently not practical for ISPs.
> > As a result ISPs don't dump the routing table or use SNMP to query the
> > routing table for debugging, but instead use other means.
> > 
> > If it could be made useful for ISPs it would make the job of writing
> > NOC software tools and ISP NMS tools all the easier.
>  
> <zathras>This is wrong tool</zathras>
>  
> To re-ask the question - drop the routing database from the MIB?
>  
> -- Jeff

All IMHO ...

If its useful for enterprise it shouldn't be dropped in the MIB just
because it is not useful for ISPs.

I think it can become ever so slightly more useful for ISPs if the
sorting order is changed.  It can also become less of a burden on
routers if the sorting order is changed.  Then any good reason to drop
it is eliminated.

At most a router can have a knob to disable it (and save the trouble
of creating the extra tree if using the old sorting order).

If some router running an OS with VM copy-on-write semantics
implements a TCP GET-BULK dump of full BGP routing table as a special
case using a fork to make the operation efficient, more power to
them.  Some ISPs might think thats great.  Others might not case as
long as it can't become a means of denial of service attack.

I think I've said enough so shutting up now.

Curtis

_______________________________________________
Idr mailing list
Idr@ietf.org
https://www1.ietf.org/mailman/listinfo/idr