How do you test maximum throughput of a switch ??

[azev]

I want to know how would you possibly saturate the bandwith of a 48Gbps backplane switch with only around 6 2ghz p4 computer ?? The reason is just for fun and for testing different brand of switch we have here.

Thanks

 

[Garion]

You really don't. You need specialized hardware to do it. Any desktop or even server platform can't push enough data and test it well enough to get any kind of valid data on switch performance.

- G

 

[azev]

Well that is sucks , I wanted to see if I can make my 2948G cisco in action with a lot of bandwidth. Currently with 35 desktop and 4 servers, it never go beyond 30% of utilization even at peak time. Well Garion, the device you're taking about, are they expansive ??

 

[spidey07]

smartbits - and they are somewhat expensive.

 

[alrox]

FYI, The 2948g seems to have a 12gbps backplane.

 

[ScottMac]

SmartBits (we use 'em in the Lab) usually figure ~US$100,000.00 per chassis (variance depends on what type of blades 10, 10/100, Gig (fiber, copper), Token Ring, DS3, OC3, 12, 48, etc ....

The price could go either way, but, generally, call it about a hundred grand per chassis (Model 6000).

FWIW

Scott

 

[mcveigh]

only 100,000! I'll take 2

 

[cmetz]

IXIA network test equipment is much cheaper than SmartBits and are more scriptable. Unfortunately, they run Windows, too. Cheaper means maybe $50k loaded

A Cisco product that has enough backplane capacity to handle full line rate is pretty rare. They're notorious for under-engineering. That said, the whole point of statistical multiplexing is that you rarely run up against your real peak need. But dropping packets, even at a low rate, is VERY bad for performance. If your max utilization is really 30%, don't worry about it, but if you ever hit 90+%, you need a better switch.

 

[cmetz]

azev, if all you want to do is send in a lot of traffic, get some dumb switches and have the PCs send out broadcast frames, then use the dumb switches' other ports to connect to the DUT switch. Effectively, you're using the dumb switches to multiply the traffic. This might even work. All you'll be able to do with that traffic is watch your port counters though, this won't help you receive the frames and look for loss that way. (port counters might or might not be trustworthy)

 

[Garion]

Originally posted by: cmetz
azev, if all you want to do is send in a lot of traffic, get some dumb switches and have the PCs send out broadcast frames, then use the dumb switches' other ports to connect to the DUT switch. Effectively, you're using the dumb switches to multiply the traffic. This might even work. All you'll be able to do with that traffic is watch your port counters though, this won't help you receive the frames and look for loss that way. (port counters might or might not be trustworthy)

Just make very sure you don't create any loops. Good way to take down your network.

- G

 

[cmetz]

Good point, let me expand a bit on it: NEVER, EVER allow any equipment in your test setup to in any way attach to your production network. (air gap!)