Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!sdd.hp.com!wuarchive!uunet!tut.cis.ohio-state.edu!mstar!mstar.morningstar.com!bob From: bob@MorningStar.Com (Bob Sutterfield) Newsgroups: comp.dcom.modems Subject: Re: PEP vs. v.32 Message-ID: Date: 1 Mar 91 17:01:32 GMT References: <1991Feb27.185340.3897@shaman.com> <3818.27ccf3f4@hayes.uucp> Sender: usenet@MorningStar.COM (USENET Administrator) Reply-To: bob@MorningStar.Com (Bob Sutterfield) Organization: Morning Star Technologies Lines: 46 In-Reply-To: tnixon@hayes.uucp's message of 28 Feb 91 12:13:40 GMT In article <3818.27ccf3f4@hayes.uucp> tnixon@hayes.uucp writes: V.32 uses echo cancellation, so both modems use the FULL bandwidth SIMULTANEOUSLY. In my simple-minded testing of various modems with our typical traffic requrements, I can attest that this is powerful stuff. Running RFC1171 PPP with RFC1144 TCP header compression over V.32/V.42/V.42bis between SPARCstation-1s with their DTEs latched at 38400bps, I do FTP transfers and check the overall and burst throughput when shipping a copy of the SPARC UNIX kernel. The bursts happen when sending those sections of the file that are particularly compressible by the V.42bis implementation in the modems. Various brands of modems don't seem to vary much in this test, so I'll cite my results for Telebit T1600s. Running one FTP, I see a sustained rate of 1.8 Kbytes/sec with gusts to 2.8. Running simultaneous FTPs (one in each direction) over the same link, they slow down only to 1.5 overall with gusts to 2.2. This is impressive in that there isn't much degradation. I don't know if the slowdown is caused by limits on the signalling capacity of the line, or by limits on the CPU and/or DSP capacity of the modems, or by limits in the Sun serial port or in the current STREAMS-based PPP implementation. (I doubt the latter, but I haven't yet run between directly-connected Suns with no modems involved.) By comparison, the same exercise between a T2500 and a TB+ running PEP with DTEs clocked at 19200 (maximum for today's PEP modems) yields a one-way throughput of around 1.3-1.5Kb/sec. I haven't tried the simultaneous two-way torture test yet, but I suspect PEP won't fare nearly so well. Between two modems running V.32/V.42/V.42bis but with one side clocked at 19200 and one at 38400, I see around 1.6 Kb/sec one-way either direction. Things still to try: 1) direct 38400 connection to see if the Sun's the limiting factor 2) bidirectional FTP over a PEP connection 3) V.32bis/V.42/V.42bis connection, at 38400 and 56K DTE rates 4) sync PPP 5) non-STREAMS, user-space PPP implementation 6) any other variables I'm forgetting? If your phone lines are clean enough, use V.32 or V.32bis modems for PPP. If you must fight dirty lines or international connections, consider using PEP. Don't use any serial line IP software without TCP header compression.