gnhlug-discuss-boun...@mail.gnhlug.org wrote on 10/01/2009 10:44:17 PM:
On 10/01/2009 08:36 PM, Bruce Labitt wrote:
Any suggestions? Places to look? Guess it is time for wireshark...
Anything else?
Are you supporting jumbo frames all the way through your stack?
No. I can't get the
I tried netpipe-tcp aka NPtcp between the nodes. Made an interesting
graph. It shows that my system doesn't exceed 100Mbps unless the message
size is greater than ~2K bytes. If the message size is over 16K, then the
network can deliver 300Mbps. At a 1MB message the throughput is ~
600Mbps.
On Fri, Oct 2, 2009 at 2:23 PM, bruce.lab...@autoliv.com wrote:
Where is the value (and what is it called) that stores the tcp buffer
size? Is it in sysctl.conf?
http://www.lmgtfy.com/?q=linux+tcp+buffer+size
-- Ben
___
gnhlug-discuss mailing list
cute...
done that search...
fasterdata.es.net/TCP-tuning/linux.html
www.psc.edu/networking/perf_tune.html
above links talk a lot about it, but don't offer the hows and why's to
find the best settings...
http://datatag.web.cern.ch/datatag/howto/tcp.html
is a little better.
Have a client-server app that is set up to communicate over a local Gbit
network. The bottleneck appears in the client to server uplink for
large payloads. At this point it isn't clear where the delay is,
(client or server side) but the sustained rate is not good enough.
Can anyone recommend
On 10/01/2009 08:36 PM, Bruce Labitt wrote:
Any suggestions? Places to look? Guess it is time for wireshark...
Anything else?
Are you supporting jumbo frames all the way through your stack?
-Bill
P.S. if you reply to a message and change the subject, the In-Reply-To
headers are preserved,