Cisco Routing Performance

mzinz

First take a look at this document:

http://www.cisco.com/web/partners/downloads/765/tools/quickreference/routerperformance.pdf

I read the top part where it mentions that these numbers are with no other services, but I'm a little confused.

It shows a 2801 handling up to 46mbps. And a 3825 handling like 150mbps. Don't these seem absurdly low? I know that a 2801 can pass traffic from g0/0 to g0/1 faster than 46mbps. Is it because all of the packets they test with are 64 bytes?

If someone could clear up what I'm seeing it would be a big help... Thanks

ConstantlyLearning

Isn't 46mbps what will be switched in hardware before being shipped off to the CPU for processing?

Doesn't mean the throughput you'll get through the router as far as I'm aware.

Met44

mzinz wrote: »

Is it because all of the packets they test with are 64 bytes?

That would be a huge part of it. They're using the chart to show the theoretical maximum packets per second -- it's not meant to be a gauge of throughput. In general, larger packets mean more bps (that's why jumbo frames are used, for instance). Lots of papers out on this if you are interested.

tiersten

mzinz wrote: »

Is it because all of the packets they test with are 64 bytes?

Its quoting worst case performance. Small packets = more overhead.

As noted at the beginning, this is for a router which has basically nothing else configured on it except basic routing from one interface to another interface. This means no ACLs, no IPS etc... Enable any of those features and the performance will drop significantly. The 1841 is rated at 38.40Mbps but Cisco only advertise it as being capable of running full services for a T1/E1. I've tried turning everything on in an 1841 and I was only able to push around 3-4Mbps through and that had some quirky behaviour regarding latency as well.

A very rough rule of thumb for software based routers used to be that you halve the rated performance value each time you enable a feature.

networker050184

You aren't going to be usually putting a pipe larger than 20mbps through a 2801 anyway. Likely you will have a small WAN link on something that small. If you want full gig line rate you are going to dropping a pretty penny on a router to do that.

ccie15672

Ignore any numbers Cisco publishes. Or Miercom. Or Juniper. Or whatever "independent 3rd party testing" company Juniper pays.

See the pattern there? Its all bullshit. Be a network engineer and actually test your configuration before hand. The good news is that you can do this reasonably well and on the cheap if you are testing low-end routers such as the ones you have mentioned.

Configure the router as it would be in production. NAT, ACLs, Encryption, etc. Then using a tool like D-ITG, iperf, or netperf, generate test traffic of varying packet sizes/ToS values/rates to see what happens. Run multiple tests each with multiple flows...

Fail interfaces. Fail routers. Fail switches. Introduce logical changes that break routing continuity (i.e., force the protocols to timeout rather than have the router force the removal of routes based on an event at layer 1 or 2)...


But never, ever trust performance numbers, feature/functionality claims, or even example configurations... Caveats and restrictions aren't always published and there is a reason for that.

shodown

ccie15672 wrote: »

Ignore any numbers Cisco publishes. Or Miercom. Or Juniper. Or whatever "independent 3rd party testing" company Juniper pays.

See the pattern there? Its all bullshit. Be a network engineer and actually test your configuration before hand. The good news is that you can do this reasonably well and on the cheap if you are testing low-end routers such as the ones you have mentioned.

Configure the router as it would be in production. NAT, ACLs, Encryption, etc. Then using a tool like D-ITG, iperf, or netperf, generate test traffic of varying packet sizes/ToS values/rates to see what happens. Run multiple tests each with multiple flows...

Fail interfaces. Fail routers. Fail switches. Introduce logical changes that break routing continuity (i.e., force the protocols to timeout rather than have the router force the removal of routes based on an event at layer 1 or 2)...


But never, ever trust performance numbers, feature/functionality claims, or even example configurations... Caveats and restrictions aren't always published and there is a reason for that.

There is a vendor out there that says they outperform vendor C in test, but they forgot to mentioned they didn't' have any of the items in bold turned on.