Date: Fri, 31 Oct 1997 01:55:01 +0000 (GMT) From: Terry Lambert <tlambert@primenet.com> To: mcgovern@spoon.beta.com (Brian J. McGovern) Cc: questions@FreeBSD.ORG, hackers@FreeBSD.ORG Subject: Re: Win 95 PPP faster than pppd? Message-ID: <199710310155.SAA07813@usr09.primenet.com> In-Reply-To: <199710302312.SAA04687@spoon.beta.com> from "Brian J. McGovern" at Oct 30, 97 06:11:59 pm
next in thread | previous in thread | raw e-mail | index | archive | help
> Anyhow, today, I ran a Windows 95 client, and a FreeBSD 2.2.2 and 2.2.5 > PPPd client through a remote access server that I'm testing. DTE rate > on the 16550s were 115200 in all cases. VJ compression on, bsd > compression off. I FTP'ed a TSB-standard file that has been rated > "very compressible". I ran dozens of iterations on both the Win 95, > and FreeBSD box, and got consistent results. > > The FreeBSD boxes managed about 8.26 K/s. The modem DTE port > was saturated at 115200bps +/- 20bps . > > The Win95 box managed about 10.5 K/s. Again, the DTE port on the modem > was saturated at 115200bps +/- 20bps. > > Anyone care to take a guess at why there is such a difference? I hand-checked > all of the transfer times and file sizes, so both clients are calculating > the throughput properly. I did notice that the FreeBSD boxes are sending > 1 1/2 - 2 times the amount of traffic back upstream (appears to be > acks from TCPDUMP on the FTP Server). Did you disable the T/TCP and piggyback ack? Don't disable piggyback ACK, and try it again. Also, does your remote access server support Microsoft compression? You are aware that you can't turn it off, only cause the negotiation to not be fatal, right? So your turning off BSD compression could mean that the BSD box is sending uncompressed and the Win95 box is sending using Microsoft's compression, negotiated on in violation of IETF standards covering PPP option negotiation. > The only other strangeness I noticed was that the window size on the > Win95 boxes were ~7-8K, compared to the 15-16K on the FreeBSD boxes. This should cause FreeBSD to send less ACK's, not more. > Just to check to see if it was a client problem, I also used fetch to > pull some files. Same results. The issue is the stack, not the program run on top of the stack. If you are saturating the serial port, the *only* thing that could account for rate differences is relative protocol overhead. Compression decreases overhead, as do piggyback ACK's. Terry Lambert terry@lambert.org --- Any opinions in this posting are my own and not those of my present or previous employers.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199710310155.SAA07813>