Zitat von Darren Spiteri <[EMAIL PROTECTED]>:
On Feb 13, 2008 2:28 PM, David Higgs <[EMAIL PROTECTED]> wrote:
Unless I'm massively wrong about what net.inet.tcp.* is used for, this
indicates that the parent was NOT testing throughput as one would
typically define it for a router/firewall. He was testing his box's
ability to send and receive TCP packets. I think these knobs are
COMPLETELY unused by the code that inspects packets and decides which
interface to send them out.
You're probably right, this has become a flame and I apologize for
being involved. I only wanted to share this info which has helped my
installations in the past.
You are right I was not not only testing throughput as it is defined
for a router. I was testing send and recieve performance because the
firewall is not only a firewall, there is a squid runninig on it too.
Out of this I assumed that the ability to send and receive TCP packets
with high performance were a pre-condition for high routing
performance. If you can not send or receive with 1GBit/s how can you
expect to be able to route with that speed? Maybe this assumption was
wrong. I will therefore do further tests with and without the
mentioned net.inet.tcp*, testing "real" throughput as well as the
ability to send and to receive.
Netperf was not the only test scenario, I was testing using netcat,
dd, ftp, scp and other things as well.
----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.
