Francesco DiMambro wrote:

> Would you be prepared to share the options you used to get the
> result, and the version of netperf? It's good to know there's more

Just the same as you. netperf -H ... -- -s 1M -S 1M
to a linux receiver (dual dual core woodcrest, barely even
noticed a 2.2Gb/s load), 1500B MTU.

The linux host was running netperf 2.4.3 (kernel 2.6.22),
and the sparc was running netperf 2.4.2 (solaris S10U4):

% netperf242 -H10.0.130.183 -C -c  -l60 -- -s1M -S1M
TCP STREAM TEST from ::ffff:0.0.0.0 (0.0.0.0) port 0 AF_INET to 
::ffff:10.0.130.183 (10.0.130.183) port 0 AF_INET
Recv   Send    Send                          Utilization       Service 
Demand
Socket Socket  Message  Elapsed              Send     Recv     Send    Recv
Size   Size    Size     Time     Throughput  local    remote   local 
remote
bytes  bytes   bytes    secs.    10^6bits/s  % M      % S      us/KB   us/KB

2097152 1048576 1048576    60.00      2239.60   4.31     4.56     5.049 
   0.667


Sendfile helps a little:

% netperf242 -H10.0.130.183 -C -c -l60 -tTCP_SENDFILE -F /var/tmp/zot -P 
0 -- -s1M -S1M
2097152 1048576 1048576    60.00      2705.02   6.45     6.03     6.252 
   0.730


(it takes a few streams to get up to line rate on a cool threads
machine :)

Anyway, if you've got an LSO implementation, I'd really like to see
LSO results compared to MDT results on the same hardware.

Drew
_______________________________________________
networking-discuss mailing list
[email protected]

Reply via email to