[c-nsp] bgp slow prefix learning

Jason LeBlanc jml at packetpimp.org
Tue Aug 21 13:51:18 EDT 2007


The slower one is also pegging the cpu, same hardware as the other.

sh proc cpu | i RIB
  99           0         1          0  0.00%  0.00%  0.00%   0 IPv6 RIB 
Redistr
 102      517640      2890     179114 60.53% 58.54% 58.39%   0 IP RIB 
Update

Jason LeBlanc wrote:
> I know one end is Cisco, a 7600/Sup720, no idea what code.  The other 
> end I have no clue.  I added the mtu, no change as of yet.  In 2.5 hrs 
> I've gotten 1087 prefixes on one, in 1 hr I've gotten 631 on the other.  
> I've never seen this move so slowly.
>
> Jared Mauch wrote:
>   
>> On Tue, Aug 21, 2007 at 12:42:14PM -0400, Jason LeBlanc wrote:
>>   
>>     
>>> I'm wondering if I have something that is by design or if something else 
>>> is slowing my routers down.  I have two 2811s with 768 megs running 
>>> 12.4(4)T3, each connected to different upstreams via fast ethernet.  I'm 
>>> using prefix lists to limit them to /22, but I should have enough ram 
>>> for a full /24 view.  My problem is that anytime I clear a neighbor it 
>>> takes them hours to learn all the prefixes again, which has forced me to 
>>> add static defaults to compensate.  My cpu is 2-4% during this, so I 
>>> can't see how its cpu bound.  Anyone have any ideas?
>>>     
>>>       
>> 	Enable "ip tcp path-mtu-discovery" globally.
>>
>> 	that should help.  If they're also using a Juniper,
>> some much older code had some odd propogation properties.  They
>> may need to upgrade code.  I think this was back in 4.x or 5.x.
>>
>> 	- Jared
>>
>>   
>>     
>
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
>   



More information about the cisco-nsp mailing list