Re: Why is my gigabit ethernet so slow?

[Martti Kuparinen <martti.kuparinen%iki.fi@localhost>]

I made some further tests...

1) No HW checksums on the router

ifconfig wm0 -ip4csum -tcp4csum -udp4csum -tcp6csum -udp6csum -tso4 -tso6
ifconfig wm1 -ip4csum -tcp4csum -udp4csum -tcp6csum -udp6csum -tso4 -tso6

2) Activate HW checksums on the router

ifconfig wm0 ip4csum tcp4csum udp4csum tcp6csum udp6csum
ifconfig wm1 ip4csum tcp4csum udp4csum tcp6csum udp6csum

3) Activated even tso4 and tso6 on the router

ifconfig wm0 ip4csum tcp4csum udp4csum tcp6csum udp6csum tso4 tso6
ifconfig wm1 ip4csum tcp4csum udp4csum tcp6csum udp6csum tso4 tso6

Results
-------
Data    Type    #1      #2      #3

 2 GB   TCP     574.12  570.93  TBD
 2 GB   UDP     869.14  867.37  867.34
10 GB   TCP     562.03  605.99  TBD
10 GB   UDP     872.86  872.75  872.80

TBDs are there because I aborted the first run after 30 minutes! So there's 
something seriously wrong with TSO as the 2 GB UDP test took about 20 sec.

As a reminder, the setup is like this:

S (Ubuntu 9.10) <--> X (NetBSD 5.99.24) <--> R (NetBSD 5.99.24)

S is sending traffic to R. The arrows are direct cables between the nodes. And 
MTU was 1500 all the time. On X and R, I have

sysctl -w net.inet.tcp.recvbuf_auto=1
sysctl -w net.inet.tcp.sendbuf_auto=1
sysctl -w net.inet6.tcp6.recvbuf_auto=1
sysctl -w net.inet6.tcp6.sendbuf_auto=1

ttcp is ran with default buffer sizes on both S and R.

Now, I wonder what kind of speed would I get with 10 Gbps cards...

Martti