On 03/28/2017 02:37 PM, Toke Høiland-Jørgensen wrote: > Matthias Tafelmeier <matthias.tafelme...@gmx.net> writes: > >> Hello all, >> >> currently, flent is only capable of fully saturating a link, or to >> what the sending node is capable to deliver, respectively. >> >> Moreover, I ran into cases where discerning finer grained saturations >> levels migh be helpful: might two digit percentage wise. >> >> I thought of introducing this with the netperf flag -W, which one can >> set the TCP SND or RCV buffer size respectively with. Netperf is >> choosing defaultwise a quite small buffer size, so that's not >> feasible. >> >> Other more primitive tools might could cover, like seeing iperf in the >> runners core as well. >> >> Not sure, though ... Netperf, as said, is not offering out of the box. >> Is there a reason? Am I running into a misconception of the usefulness >> of such a feature? Maybe fully-saturated|partly-saturated|non-loaded >> are the states telling all? > The problem is that specifying a particular bandwidth with TCP is non > straight forward. Sure, you can cap the window (I think the reason > netperf starts out small is that the kernel will auto-tune it by > default), but the actual throughput achieved will then depend on the > RTT. So you can't reliably say "give me X Mbps of TCP throughput". You > could do it with UDP (and there's a Flent test that will do that), but > then you just get a burst with no feedback, and only a single throughput > value at the end. Am aware of all of that. No, no, I thought I could clamp the auto tuning socket wise! But perceive all of that rather as ramblings of despair. It's the past.
> As for the usefulness... well, what's your use case? :) Don't want to burden anyone further! The point is, I've found links and scenarios where scaling by flow numbers is more than granular enough for my needs. Sorry for not keeping dated. Many thanks for coming back anyhow! -- BR Matthias Tafelmeier
0x8ADF343B.asc
Description: application/pgp-keys
signature.asc
Description: OpenPGP digital signature
_______________________________________________ Flent-users mailing list Flent-users@flent.org http://flent.org/mailman/listinfo/flent-users_flent.org
