thrulay-users - Re: thrulay results over ultralight
Subject: Discussion list for thrulay, a network testing tool
List archive
- From: stanislav shalunov <>
- To: Yee-Ting Li <>
- Cc: "Connie A. Logg" <>, Les Cottrell <>,
- Subject: Re: thrulay results over ultralight
- Date: 19 Oct 2005 23:32:03 -0400
Yee-Ting Li
<>
writes:
> #(ID) begin, s end, s Mb/s RTT, ms: min avg max
> ( 0) 0.000 5.001 31.334 82.304 3405.275 4934.103
> ( 0) 5.001 10.001 93.968 5015.022 7633.145 9678.344
> ( 0) 10.001 15.001 146.852 10035.642 12392.504 14508.290
> ( 0) 15.001 20.001 175.368 14858.923 17195.312 19471.865
> ( 0) 20.001 25.000 199.296 15123.740 17989.805 20905.669
> ( 0) 25.000 30.001 218.963 13345.205 14168.544 15325.002
Yee,
It looks like you're seeing the dreaded Linux internal buffering.
(There's no good reason Linux should buffer 10-15 seconds worth of
stuff, but sometimes it does. When this happens, TCP also becomes
more sluggish---perhaps there's a queue scan somewhere inside the
kernel, but that's just a guess.) Thrulay reveals it, but doesn't
cause it. Can you try running a single stream with window sizes of
10-25-50-100MB (so that the window size is more in line with
throughput*delay)? If it does better, it might help to reduce
IP-level queuing and/or driver buffers.
--
Stanislav Shalunov http://www.internet2.edu/~shalunov/
This message is designed to be viewed at sea level.
- thrulay results over ultralight, Yee-Ting Li, 10/19/2005
- Re: thrulay results over ultralight, stanislav shalunov, 10/19/2005
Archive powered by MHonArc 2.6.16.