This is something I've said multiple times in the past about network switches - provided you get a remotely decent one, it won't have an appreciable impact on performance regardless of how many ports are utilised simultaneously. Of course, if you have for example two input ports fighting over a single output port, then they'll have to share the bandwidth, but I'm talking about total switching throughput here.
A few people have disagreed with me, saying basic switches won't allow full throughput and/or only high-end switches will do so. This goes against the 'non-blocking switching fabric' system these switches claim to use, but I didn't really have any evidence besides a few tests I've performed myself with just two streams to support my claim.
I have been meaning to do some more extensive testing myself but I'd need access to something like eight systems capable of gigabit throughput, and probably running Linux, as iperf for Windows is a bit strange. And I don't, so I haven't.
However, this article just caught my attention, and shows the switches are indeed capable of wire-speed performance on multiple ports simultaneously. In fact, the n stream tests are almost exactly n times the 1 stream test!
http://uk.hardware.info/reviews/4795...cp-performance
Oh and I'm not trying to say all switches are equal, because the story obviously gets more complicated when you start involving managed switches/rule sets/VLANs/etc.


LinkBack URL
About LinkBacks
Reply With Quote