On Jun 11, 2011, at 9:53 AM, Lars Noodén wrote: > On 6/11/11 6:51 PM, RB wrote: >> On Sat, Jun 11, 2011 at 08:27, Lars Noodén <lars.noo...@gmail.com> wrote: >>> Would that mean that a pair of machines transferring at full speed on >>> eth0 and eth1 would not reduce the available bandwidth for another pair >>> on eth2 and eth3? If not, then how can that be achieved. >> >> As has been noted a couple of times before on this list (indeed, on >> threads you've started), software switching is slow and you're not >> going to get anywhere near the performance of a hardware switch, >> especially with embedded hardware. Your bandwidth is limited by your >> processor and memory, not your port speeds. > > Thanks. It would be nice to have a writeup of something like that, with > how and why, in the Wiki: > http://wiki.soekris.info/
it's four 10/100 ethernet interfaces in a pc, in this case on a common 32 bit share pci bus... so it shares a common sets of properties with most other pc's and all linux or *BSD boxes that use the same kernel subsystem for networking.... >> That is: it doesn't matter if you have 4 GbE interfaces if your >> processor/memory/kernel can only achieve, say, 1Gb/s. If two ports >> start consuming 800-900Mb/s, there's only 1-200Mb/s left for other >> ports, and there's little you can do to prevent that without taking >> bandwidth away from the initial two. > > How would that be best measured? a 32bit 33mhz pci bus will carry at most 120MB/s the reality is generally somewhat lower than that. the easiest way to conclude how much network throughtput you can achieve is in fact to test it. > /Lars > _______________________________________________ > Soekris-tech mailing list > Soekris-tech@lists.soekris.com > http://lists.soekris.com/mailman/listinfo/soekris-tech > _______________________________________________ Soekris-tech mailing list Soekris-tech@lists.soekris.com http://lists.soekris.com/mailman/listinfo/soekris-tech