[c-nsp] Growing BGP tables

[Robert Boyle]

At 10:33 PM 11/24/2004, you wrote:
>Maybe there are some Livingston engineers around who did BGP on the
>Portmaster3 that can give some ideas on memory usage.
>  From what I remember the PM3 was able to fit two full views (back
>then about 100k routes each) into 32MB of RAM.
>
>The minimum amount of space needed to store 140k routes is 1.6MB (4
>bytes for network, 4 bytes for netmask, 4 bytes for destination).
>The overhead per route must be huge.

Actually, Livingston/Lucent had one guy who was a contract programmer who 
developed their BGP implementation in a very short period of time. It had 
one minor bug which he fixed quickly and there were never any additional 
updates to it after that. We ran BGP on a PM3 with 3 T1 lines (using the 
infamous PM-SYNC-T1 card for the 3rd T1) for several months with 3 full 
feeds back when we had about 55-60k routes and we always had about 20MB 
free with 32MB in the box. It worked well and was very stable. I don't 
recall his name, but he was a very smart cookie. He also wrote their OSPF 
code too iirc. This is what I was told by my inside people at the time 
anyway. I think there is a lot which can be done to extend the life of 
256MB routers with better table/RAM management as is being discussed. Does 
Cisco have enough incentive from the stick of many unsatisfied customers to 
do this or is the carrot of lots of NPE-400 and G1 and soon G3 cards (for 
7206s anyway) a more powerful motivator? However, if we have to do a 
wholesale forklift upgrade across all of our core and edge routers, we will 
be looking very hard at other vendors with arguably better performance and 
code stability when it comes time to upgrade.

-Robert


Tellurian Networks - The Ultimate Internet Connection
http://www.tellurian.com | 888-TELLURIAN | 973-300-9211
"Good will, like a good name, is got by many actions, and lost by one." - 
Francis Jeffrey