[c-nsp] 2611XM throughput

[Justin Shore]

Adam,

I had an identically configured 2611XM in service a couple years ago. 
The CPU maintained 70-90% during peak times.  At the time that was with 
about 8Mbps passing through it.  One night when the router reached just 
under 10Mbps the CPU pegged at 100% (usually I only see Cisco IOS 
devices report 99% but this one said 100%) and fell off the network (I'm 
assuming that OSPF timed out).  The ensuing flood of packets from behind 
the router (cable modem network with about 400 users) easily exceeded 
the pps of the router and kept the CPU pegged generating ICMP 
Destination Unreachables.  Even consoling into the poor thing was 
pointless as it took a couple minutes to output a single character to 
the CLI.  It had 2 ACLs of about 40 lines combined and was running OSPF 
with one neighbor (50 routes).  I had a replacement 2821 overnighted to 
me the next day and I replaced the 2611XM that night.  The performance 
difference was very noticeable, even with the same approximate 
throughput.  The 2611XM introduced significant delay.  IMHO the 10Mbps 
ceiling is absolute if not exceedingly generous.

The 2611XM is good as a CE with a couple T1s.  Beyond that toss it on 
your pile of 2500s and go buy an ISR.

Justin


Adam Greene wrote:
> Hi,
> 
> I'm trying to determine what kind of throughput we can expect from a 2611XM currently in production (IOS 12.3(24), 128MB RAM). The router is doing eBGP to two peers but only advertises one network and receives default routes only. Other than that it's plain vanilla.
> 
> Cisco's router performance guide rates this router at 10.24Mbps (is that one direction at a time or bidirectional, I wonder) with an avg packet size of 64bytes. Judging from the # of packets and the # of bytes transferred on the unit's WAN interface, I'm calculating average packet size at about 154bytes per packet. 
> 
> Based on these specs, would it be realistic to expect the router could push about 25Mbps aggregate (for example, 12.50Mbps in both directions simultaneously)?
> 
> Thanks for your insight.