On Feb 12, 2008 8:37 PM, raven <[EMAIL PROTECTED]> wrote: > Ted Unangst ha scritto: > > On 2/12/08, Darren Spiteri <[EMAIL PROTECTED]> wrote: > > > >> I don't know why or how this poorly documented sysctl works, but the > >> result speaks for itself. Note the dramatic throughput increase of the > >> parent. > >> > > > > running netperf on a firewall is a poor test of forwarding performance. > > > > > > > So, Ted, what test he need to do, to have some results good ? I ask > because i was thinkin to use netperf too to test my network performance, > so if you can help both of us, i really appreciate. > Thanks.
What's your definition of "network performance?" I believe Ted's point is that receiving and sending packets (i.e. using it as an endpoint) is the job of a server, not a router/firewall. Routers and firewall performance is measured in how well they can inspect and forward packets that are NOT destined for the hardware under test. That's probably why sendsize/recvsize should not matter on a router/firewall; I think these knobs configure how much space is allocated for endpoint send/recv operations. Can you use netperf to send traffic THROUGH the box you want to test? Otherwise, generate traffic load through your box using applications like wget, scp, ftp, tcpreplay, or a traffic generator. If you only care how much traffic your box can SERVE, you should describe it as a server. Here, netperf may be more suited to measuring your performance, or you can point the previously mentioned tools at your server. --david
