[isp-bgp] Re: Smaller ISP, do we need full BGP routes?

[sjk@...]

On 31 Jul, Anthony Guida wrote:
> Hello,
> 
> I work for a smaller ISP in Chicago and we are running
> a multi-homed BGP environment with 2 backbone
> providers. Our edge router is a 7206VXR (NPE300) with
> 256MB of RAM. We have 6 T1s plugged into this router,
> 3 to each backbone provider. We are also currently
> taking full BGP feeds from both providers.
> 
> I have been noticing somewhat of an increase in CPU
> and memory usage on the edge router in the past few
> months and would like to get some assistance on
> whether or not we should be taking full BGP feeds
> since I understand this takes up a lot of CPU and
> memory. Another reason for this performance/latency
> throughout our network.
> 
> According to our MRTG graphs, I can say 99 percent of
> our traffic is inbound. Since we are an ISP, everyone
> is downloading images and files more than uploading
> like a web hosting provider would. What I mean is
> taking full BGP feeds for outbound traffic really
> shouldn't matter. We should just take partial or
> default routes and let our backbone providers decide
> where to send the outbound traffic. Our inbound
> traffic is determined through our AS prepending &
> advertisements config.
> 

As I am sure others have mentioned, if most of your taffic is inbound
the you probably don't need full views. However, a 7206 w/ 300 NPE
should have no problem whatsoever handling this. Thus, I think you've
mis-diagnosed the problem. One of our peering routers is a 7204/300 npe
-- it handles two views one from a DS3 another from one of our RR - it
normally sits around 10% CPU utilization. Some things you may want to
look at -- if your CPU is higher then say 25%. . . Make sure CEF is
enabled and packets are not moving to slow path switching; How are you
IMUXing your T1's? I've seen problems with PPP multilinks having
incrementing encapsulation errors -- do a show ppp multilink -- this
usually accompanies high latency across the link. I have also seen UDP
floods cause high cpu utilization -- and we've gotten to a point where
we apply CAR limits to outbound UDP traffic before it is forwarded to
the edge.

Hope this helps -- sjk
  
---- Audax et Cautus -------



------------------------ANNOUNCEMENT---------------------------------
---------------------------------------------------------------------
           ISPCON FALL 2003 - Santa Clara Convention Center
     October 20-22 - The Definitive Event Wired and Wireless ISPs
                            www.ispcon.com
----------------------------------------------------------------------
----------------------------------------------------------------------