Routing performance between two subinterfaces in a 3640-router?

[amielonen]

We have a 3640-router with one NM-2CE1B and one NM-1FE-TX module,

and running IOS c3640-is-mz.121-16.bin.

We have made a network performance test through the router,

between two hosts on separate VLANs, using the TTCP-tool for Windows.

http://www.cisco.com/warp/public/471/ttcp.html

(We have also test with the Netperf-tool, www.netperf.org)

But we can only pass the router at approx. 40 Mbit per second.

(We have also done this test in a experimental environment,

another 3640 and only 2 host with the same result.)

The NM-FE1 is connected to a Cisco 3548-switch, ISL trunking and

of course 100 Mbit /Full duplex at each side.

We have tested with combinations of 'ip cef', 'ip route-cache flow/same-interface'

and so on.

Is 40 Mbps a proper result when routing between two subinterfaces

or is there any way to increase the performance?

Best Regards,

Anders

[DALE FRANCIS]

So you are doing a perfomance test from one media to another. This has a potential to slow the box down, i have found some figures which are quoted by Cisco. The bottle neck in your scenario is defs the CE1.

For the NM-2CE1B:

The total aggregate throughput performance of the 2 port primary rate interface (PRI)/CE1 network module is 16 kpps. Use Cisco IOS CPU-intensive features such as Multilink and compression with more than 1 PRI configured with caution if traffic loads above 40 kbps per channel are expected.

For the NM-1FE-TX:

With 64 byte IP packets: 60 kpps full-duplex and 30 kpps for inter-switch link (ISL).

Basic Performance Charts:

Cisco 3640 100MHz RISC 250 Mbps

Fast Switching: 50 - 70 kpps

Process Switching: 4 kpps

[amielonen]

Thanks for the answer.

We are not doing this test between two different media.

The CE1 is not involved at all.

It´s only IP routing between two subinterfaces with ISL encapsulation

at the 1FE-TX module, connected to a Catalyst 3548-switch (100 Mbit FD)

Fast switching is on.

Is this asuming correct?:

With 64 byte IP packets: 30 kpps for inter-switch link (ISL) should give

a speed of (64x8) x 0.030 = 15.4 Mbps (which is not so impressing...)

But with another size of IP packet?

I mean, 1300 byte gives 312 Mbps, a impossible result.

Which size do Cisco mean in their "kilopacket per second" tests?

Regards,

Anders

[DALE FRANCIS]

Okay... just that you mentioned the the CE1 which confused me a bit. :)

To the best of my knowledge they use 64byte packets in the testing... and i know the theory behind the increase in performance on CCO, this is a result of total throughput due to larger packet but with no timing contraints.

[dfelter]

I've run performance tests using SmartBits between 2 3620s each with an NM-1FE-TX and interconnected via NM-1A-OC3MMs (OC-3 ATM modules). I was never able to achieve better than 28 Mbps at 1518 bytes. I imagine that with the faster processor, 40 Mbps is the performance limit of the 3640; 3600s are not "wire-speed" routers.

What is the CPU utilization of the router when performing these tests? I'll bet that at 40 Mbps aggregate throughput, you're seeing close to 100% utilization.

[bourse]

I believe the model number has something to do with the router performance. From discussions that I have had with my Cisco engineer, a 3620 can handle roughly 20mbps, a 3640 approx. 40mbps, and a 3660 approx 60mbps...