|
From: | Goswin Brederlow |
Subject: | Re: [Mldonkey-users] Re: Perfect amount for used sockets found |
Date: | 30 Jan 2003 10:59:11 +0100 |
User-agent: | Gnus/5.0808 (Gnus v5.8.8) XEmacs/21.4 (Military Intelligence) |
Fabian Henrici <address@hidden> writes: > Sven Hartge wrote: > > > > > >>>>>At 250 I achieved about 5 kB/s. > >>>>>At 150 about 6.5 kB/s. > >>>>>At 100 about 7.0 kB/s. > >>>>>At 50 about 6.6 kB/s. > > > > > > At least I can reproduce those figures at will. So it is rather > > unlikely > > > those figures are random or induced by external influences. > > YMMV for bigger lines though. > > > S° > > > > > > Hi, > > I tried different values for a 768/128kbit DSL line and found 275 to > work perfect. But I think values between 250 and 300 should be tested. Could you gather statistics about packet size too while you test? Linux tries to delay small packets in tcp streams in the hope that more data will follow. I would think the more sockets mldonkey spreads it upload bandwith over the less successfull this becomes and more overhead due to tcp/ip headers is produced and thus less speed. A packet distribution by size graph could indicate such behaviour. MfG Goswin
[Prev in Thread] | Current Thread | [Next in Thread] |