On Mon, 24 Sep 2001 21:12:16 BST, Matthew Sackman writes: >Ah ha! <normally I´d snipped that, but..oh, so nice ;-) > >Things are not as bad as they seem. > >If I do cat /dev/zero | nc -u -p 1666 doris.namkas 1666 >and set up doris.namkas to receive correctly then knetload reports >around 98000KBit/s which doesn't seem too bad. > >Doing it the other way round reports the same (ish, though in this >case, the sending machine - doris - is only a PII and is also >compiling gcc at the same time!). /dev/zero consists of just a stream of 0x00´s, not that cpu-intensice per se. >Doing both at the same time (different ports) gives the first host >(a PIII) to send at about 100,000KBit/s and doris to send at about >25KBit/s - it doesn't matter who starts - doris always sends slower >it seems. What about the receiving rate? Could be a full-/half-duplex issue..or a "switching hub" one[0] or... >This is the first time that I've done testing like this - does >anyone have any comments to make about what this shows or doesn't? Try it w/ tcp (eg without "-u"´s), too. And cease compiling wile testing ;-) Post results. Try with both the supplied drivers and with the plain scyld.com ones. Post results. Let tests run for at least 15 minutes each. Post results (Of course, you could do the mentioned and *summarize* the results ;-) ). And, if for my sake only, use iptraf, which runs without any X-GUI, eg also on my slower boxen, I have more faith in such ;-) 0: Anyone fit for a proper rant about that crap? cheers, &rw -- -- Benutzerfreundlichkeit -- Der Benutzer hat zum Admin freundlich zu sein. (Thorsten Fenk) ----
Attachment:
pgpy9fWi9rR60.pgp
Description: PGP signature