Looking for switch recommendations ...
bakul at BitBlocks.com
Fri Mar 26 12:54:21 PST 2004
> > For 100Mbps ports, the max packet rate in one direction is 10^8/672 ==
> > 148809 pps (packets per sec) per port. So for 24 port full duplex ports
> > you get an aggregate maximum throughput of 148809*24*2 = 7738068 =
> > 7.14Mpps (Million pps). For a 48 port switch it is 14.29Mpps.
> so, the closer the Mpps gets to that 7.1Mpps, the better the switch
> overall? I take it that has to do with the CPU driving the switch itself,
> or is there other factors that help drive that # up?
Well, "better overall" involves a lot more than the max
throughput -- you are very very unlikely to see nothing but
64 byte pkts in your network (typically 50% of pkts are acks,
the other 50% are MTU size) so 6.6Mbps seems good enough to
me. I would look at quality first, and then service, how
this switch is to be used and whether there are other
features that may be relevant (such as vlan, QoS etc). Also,
I would choose a switch to ensure there is about 50% to 100%
headroom for growth (in # of ports, etc).
In terms of achieving max throughput it depends on how the
switch is engineered. CPU driving the switch matters less
than whether they have done a balanced design. You also need
a backplane fabric that is capable of delivering full b/w no
matter what the traffic pattern is (you may not achieve
7.14Mbps for a given pattern but for each pattern one can
calculate a theoretical max which should be achievable).
In any case, quality of the switch construction probably
matters the most. If rj45 ports are flimsy nothing else
More information about the freebsd-net