[nsp] Throughput of 100 Mbps

Vincent De Keyzer vincent at dekeyzer.net
Fri Nov 7 05:10:12 EST 2003

You are raising a very interesting question. I never understood (or
seriously tried to understand) what is meant by this throughput numbers.

Are you saying that the pps numbers from Cisco are always for 64-bytes
packet? Could be, in fact this gives the highest pps number... And it's
always one-way, right?

The test I made was very simple: a 3620 with 2 FE interfaces, and a PC on
each of those; then FTP between the two PCs.

I only had 40% of CPU usage - why isn't it trying harder, to reach a higher
throughput? :-) 


> > what's the smallest (read: cheapest :-) router that can
> > process 100 Mbps of traffic?
> A 3745 per a Cisco rep.  But I don't know if they just gave 
> me the "official answer" by looking at the pps rating and 
> running the numbers with 64-byte packets (225,000 * 64 = 
> ~110Mbps), or if the statement applies to full-sized packets 
> as well (e.g., from testing that they've done in-house).
> The reason this matters is because I was able to get 40Mbps 
> out of a 2621 with full-sized packets using TTCP.  The 2621's 
> official pps rating is 25,000, which with 64 byte packets is 
> ~12Mbps.  So the bandwidth you can get with full-sized 
> packets is, not surprisingly, significantly higher than with 
> minimum-sized packets.  If this is the case for the 3x00 
> series as well, there's hope that something cheaper than a 
> 3745 may be able to pull off 100Mbps will large packets. 
> (Note however that it's actually 200Mbps of traffic that 
> saturates a full-duplex FE interface, if that's a concern in 
> your case.)
> Do any of the Cisco people that hang out here have any 
> information on this?  I've asked this elsewhere and haven't 
> had any luck getting an answer.
> > Playing with a 3620, I could only get around 40 Mbps. What
> > next? Do I need 7200? What about the 2600 XM, which I know 
> > very little about?...
> Given what I got with a 2621 I would have thought a 3620 
> would be able to do somewhat better, but perhaps not.  What 
> kind of test did you run to get your number?
> -Terry

More information about the cisco-nsp mailing list