NetNews Usenet Archive 1992 #31/NN_1992_31.iso/spool/comp/dcom/modems/18777

Internet Message Format | 1993-01-01 | 3.6 KB

Xref: sparky comp.dcom.modems:18777 comp.sys.sun.hardware:6587 comp.sys.sun.admin:9859
Newsgroups: comp.dcom.modems,comp.sys.sun.hardware,comp.sys.sun.admin,cu.sun.managers
Path: sparky!uunet!zaphod.mps.ohio-state.edu!news.acns.nwu.edu!nucsrl!ddsw1!karl
From: [e-mail] (Karl Denninger)
Subject: Re: Anyone out there successfully get a 14.4K modem running on a Sun?
Message-ID: <[e-mail]>
Date: Fri, 1 Jan 1993 22:37:58 GMT
References: <[e-mail]>> <[e-mail]> <[e-mail]>
Organization: MCSNet, Chicago, IL
Lines: 62

In article <[e-mail]> [e-mail] (Larry Snyder) writes:
>[e-mail] (Bob Sutterfield) writes:
>
>>Using the WorldBlazers' V.32bis carrier, you'll see FTP throughput
>>around 2.4 Kbytes/sec for /vmunix, or 2.8 Kbytes/sec for text files (I
>>use 5 concatenated copies of /usr/dict/words).  Text files go faster
>>because they're more susceptible to in-modem V.42bis data compression.
>>An 8-byte ICMP ping packet shows a minimum round-trip time around
>>240ms, and a single-character telnet packet shows a minimum round-trip
>>time around 150ms.
>
>Have you played with other modems like the USR Courier v.32bis modems?
>Folks (including myself) have found a reduced latency with these modems
>improving interactive throughput.
>
>Do you suggest MNP enabled on SLIP v.32bis lines -- or just v.32bis
>and v.42bis?

Use V.42bis/LAPM!  MNP 5 can be a serious lose on compressed data!

You're doing something wrong if your best text numbers are under 3KB/sec
(your TCP connection window is almost certainly too small).

With proper window settings on both ends, and VJ header compression, I am
getting ~35Kb/sec on text FTPs over a V.32bis link running PPP.  This is
with >cheap< modems (Supras) on one end, and T3000s on the other.  If I
force the Supras out of the loop so I'm running T3000 <> T3000 I can
achieve nearly 5Kb/sec >reliably< on most text files over 100K or so.

A "get" of /vmunix returns much slower rates, of course.  There the modem
difference is not nearly so profound.  My average is in the ~2.7Kb/sec
range.  /usr/dict/words goes at a solid 3.3kb/sec rate.  Note that
/usr/dict/words isn't all that typical a test; it doesn't have much
whitespace in it.

The ping turnaround time (~220 - 240ms) quoted above is reasonable but not
great.  T3000 <> T3000 I get around 180ms for 8-byte ICMP messages, and ~240ms
for 56 byte messages.  I would be quite interested in trying a couple of other
modems, however, I don't have any others ;-)

This is with Netblazers on both ends, going to and from an Ethernet host on
each side.  The ethernet propagation delay + host time is a few ms at most.

A TCP window size large enough to guarantee that he channel doesn't starve
waiting for the packet ACKs is >essential< to proper performance.  Without
this you're wasting a significant part of your capability.

Telebit's Netblazer defaults (2048 for the TCP window), and I suspect many
others, are far too low for optimum performance with today's V.32bis modems.
I would hope that Morningstar lets you tune these parameters (particularly
window size and MSS); if not you're going to have to accept what you've got
now.

One word of warning -- setting these parameters up higher >can< affect
Netblazer (or other router) memory usage severely as packet buffer memory
consumption will rise dramatically.  This might be bad if you are tight
on RAM.

--
Karl Denninger ([e-mail], <well-connected>!ddsw1!karl)
Data Line: [[phone]] Anon. arch. (nuucp) 00:00-06:00 C[SD]T
Request file: /u/public/sources/DIRECTORY/README for instructions