[c-nsp] c3560 L3 throughput with BGP

[Valentin Stoicescu]

Adrian Minta wrote:
> Jon Lewis wrote:
>   
>> On Mon, 29 Oct 2007, Adrian Minta wrote:
>>
>>     
>>>> 1. Handling approximately 500Mbps each way (about 1Gbp total 
>>>> throughput)
>>>> 2. Handling BGP peerings with up to 20 peers with total of 200-300 
>>>> prefixes
>>>> 3. Complete BGP support (communities, filtering etc.)
>>>> 4. Of course boxes are basically acting as router, so both ports are
>>>> separate L3 ports.
>>>>
>>>>         
>>> C3560 and c3750 are both switches witch means line rate speeds as long
>>> as your number of routes is smaller than TCAM size. One prefix more and
>>> everything goes kaboom.
>>> I suggest to take a look at Cisco 7201 router.
>>>       
>> Why?  300 routes on a 3560 is nothing.  I've done around 1300 routes 
>> and 2-3 BGP peers on a 3550.  20 peers seems like maybe a lot, but as 
>> long as they're all reasonably stable, I suspect it'd work and be much 
>> cheaper than a 7201.
>>     
> Are you sure that nobody is accidentally injecting more routes ?
> You will always be on the edge. C3550 was a better switch than C3560 IMHO.
>
>   
If the max prefixes received doesn't exceed 8000 unicast routes with sdm 
prefer routing  3560 it's capable of dealing with all that bgp peers. 
I'm currently running 2 bgp sessions with a total of bandwidth of 
200Mbps and ~3400 prefixes recived from each peer
##sh ip bgp su
4051 network entries using 473967 bytes of memory
BGP using 1401759 total bytes of memory
Neighbor        V    AS MsgRcvd MsgSent   TblVer  InQ OutQ Up/Down  
State/PfxRcd
81.x.x.x           4      X   56484   33871    76815    0    0 
3w2d         3394
89.x.x.x           4      X   90394   33871    76815    0    0 
3w2d         3477

#sh proc cpu | i CPU
CPU utilization for five seconds: 23%/0%; one minute: 22%; five minutes: 22%
And if a peer is stupid enough to flood you with prefixes there is 
neighbor maximum-prefix.