Re: network speed

[Matthias Scheler <tron%zhadum.org.uk@localhost>]

On Tue, Jul 21, 2009 at 03:21:09PM +0100, Patrick Welche wrote:
> quantz% ttcp -r -s -b 131072 -u
> ttcp-r: buflen=8192, nbuf=2048, align=16384/0, port=5001, sockbufsize=131072  
> udp
> ttcp-r: socket
> ttcp-r: rcvbuf
> ttcp-r: 67018752 bytes in 14.89 real seconds = 4394.53 KB/sec +++
> ttcp-r: 8183 I/O calls, msec/call = 1.86, calls/sec = 549.45
> ttcp-r: 0.0user 0.0sys 0:14real 0% 0i+0d 0maxrss 0+1pf 8172+0csw

You should really use "-l 1472" to avoid IP fragmentation.

> simple question: why is the number of bytes different in that transfer
> according to the receiver and the transmitter?
> (nearly exactly factor 4 difference)

Because UDP doesn't support flow control and a lot of the packets
got dropped as a result.

        Kind regards

-- 
Matthias Scheler                                  http://zhadum.org.uk/