Activity Summary
- 6870 days inactive
- 6870 days old
- pacnog@pacnog.org
- 1 participants
- 0 comments
j
: Next unread message k
: Previous unread message j a
: Jump to all threads
j l
: Jump to MailingList overview
Hello,
I'm just forwarding this in case its of interest.
Save.
---------- Forwarded message ---------- Date: Thu, 2 Dec 2004 14:39:01 +1000 From: David J. Hughes bambi@Hughes.com.au To: Oz-ISP Mailing list aussie-isp@aussie.net, isp-australia@isp-australia.com Subject: [Oz-ISP] IOS and Growing BGP tables
Hi,
I'm posting this in an attempt to raise a bit of support. I've been arguing for a feature enhancement for IOS that will prolong the life of most Cisco routers that take a full BGP table. The growth of the table is causing concern for many major providers who can see the RAM on their lower end 7200's running out in the near future. Their default solution is not to upgrade the boxes, but to filter out the /24 advertisements.
However, for many years in Australia, a /24 was the normal allocation. If they blindly filter out all the /24's then lots of networks in Australia (and other countries) will become unreachable. The proposed features allows you to filter a prefix if there is a less specific prefix already in your table (with various tuneable knobs with respect to next hop etc).
Some guys at Cisco have taken this on and filed a bug / feature request for this. The more people that open a TAC case and have it linked to this bug ID the higher the priority of this request will become. It's being supported on the Cisco-NSP mailing list but I thought it worth "rattling the cage" on these lists as well. The more voices the better.
The Cisco Bug ID is "CSCsa45474 : Ability to block overlapping BGP prefixes from being installed in RIB". Some history of the discussions is included below. If you have a TAC account and are so inclined, please open a new case and ask the engineer to link it to this bug ID.
Thanks
Bambi ...
Begin forwarded message:
From: David J. Hughes bambi@Hughes.com.au Date: 1 December 2004 10:26:18 AM To: 'cisco-nsp' cisco-nsp@puck.nether.net Subject: Re: [c-nsp] Growing BGP tables
I hope that most people on this list would open a case to try to get a feature like this included in upcoming IOS releases. It's about the only thing that will extend the life of your "RAM limited" routers without breaking connectivity to historically legitimate /24 allocations.
David ...
On 01/12/2004, at 9:40 AM, Rodney Dunn wrote:
For those that have the ability to open a TAC case and they think this is something they could use, open a case and have the TAC engineer attach the case to this bug:
CSCsa45474 Externally found enhancement defect: New (N) Ability to block overlapping BGP prefixes from being installed in RIB
------
Begin forwarded message:
Anything that reduces the consumption of RIB and FIB resources as a result of de-aggregation etc would be a very welcome step in the right direction. Thanks for looking into this Rodney.
David ...
On 24/11/2004, at 3:33 AM, Rodney Dunn wrote:
My point was that you can't do it with BGP because you don't have a way to request a re-advertisement from a peer when the less specific goes away.
There is a possiblity we could keep the more specifics out of the RIB which would save memory there and in the FIB.
That being said the best that can be done currently appears to block the routes from going in the RIB. We are looking to see how complex that would be to do.
Rodney
On Tue, Nov 23, 2004 at 11:11:52AM -0600, Brian Feeny wrote:
I like this, basically all your doing is getting rid of the redundant information which serves no purpose other than taking up space in the RIB/FIB anyways. And if someone should further deaggregate or make a change, then the new update is going to run its algorithm against the current RIB all over again.
On Nov 22, 2004, at 4:16 PM, David J. Hughes wrote:
Hey Rodney,
How about ...
for each prefix received in update for each more specific prefix in RIB if RIB prefix ge /24 and same next hop as update prefix drop RIB prefix if update prefix ge /24 for each less specific prefix in RIB if RIB next hop same as update next hop drop update prefix
something like that would filter during the update and wouldn't require another full copy of the table to be held. I'm sure you guys could come up with something more efficient but the basic idea would work wouldn't it?
David ...
---- email "unsubscribe aussie-isp" to majordomo@aussie.net to be removed.