How We Test Hardware Routers: Revision 10

[thiggins]

Our Revision 10 Router test process brings a renewed focus on performance.

Read on SmallNetBuilder

 

[angel hsu]

hi thiggins,
I setup a HTTP score test environment, but i don't understand how to get a "score"? like ASUS Lyra are scores of 1.6, 3.5, 23.7 and 56.3 for 2KB, 10KB, 108KB and 759KB file sizes.
i think these scores get from calculate some values . Could i know how to get it?

 

[thiggins]

The HTTP score calculation is described in the article.

Scores are expressed as percent of routed throughput vs. throughput of client and server directly connected. Maximum score of 100% means routed throughput was equal to direct-connect throughput.

The direct-connect throughput values are shown in the graph in the article.

 

[angel hsu]

Thanks for your replay, i try to duplicate, but the results are different with your. I will continue try it. Maybe the parameter of apachebench use wrong.

 

[thiggins]

Use the settings described here.

Ethernet NIC performance is also important.

 

[angel hsu]

Thanks thiggins, I get very similar result.

 

[thiggins]

what did you change that made the difference?

 

[angel hsu]

follow the blog, I change ApacheBench's parameter and change an Intel NIC thanks for your help!

 

[System Error Message]

but will it route?
How about when testing 10Gbe?

 

[coxhaus]

How about 2 one gig connections. This will test load balancing and the load should be greater than most of the routers can handle.

 

[coxhaus]

I thought of something else is in your test you are excluding DNS. It sure would be nice to include it. I think one of the problems I had was DNS slowing down web pages. DNS should have the highest priority. But I think I got caught with DNS running slow.

 

[thiggins]

How about 2 one gig connections. This will test load balancing and the load should be greater than most of the routers can handle.

Sorry, I don't understand this suggestion.

It's unlikely I'll expand testing to include DNS caching.

 

[sfx2000]

Ethernet NIC performance is also important.

with 1Gbe - all NIC's perform about the same - some will suggest that Intel NIC's are somehow better...

In my experience - Intel might have an advantage across different OS's (Intel works really nice on *BSD and Solaris), but Realtek is more than sufficient if no problems out of the box.

 

[System Error Message]

Use the settings described here.

Ethernet NIC performance is also important.

with 1Gbe - all NIC's perform about the same - some will suggest that Intel NIC's are somehow better...

In my experience - Intel might have an advantage across different OS's (Intel works really nice on *BSD and Solaris), but Realtek is more than sufficient if no problems out of the box.

not quite, all platforms are the same, but different NICs differ. For instance realtek could beat marvell or intel if you have a fast CPU, and cable tolerances for example, not to mention variations within the same brand. To eliminate this problem, you first do a direct test between PCs by connecting them directly and running the tests.

It doesnt matter what NICs you used, once you set them up, do a direct test and keep to those settings when testing routers. Someone also was actually asking about testing dual WAN.