home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.dcom.modems
- Path: sparky!uunet!cis.ohio-state.edu!mstar!mstar!bob
- From: bob@MorningStar.Com (Bob Sutterfield)
- Subject: Re: PEP over SLIP?
- In-Reply-To: vjs@rhyolite.wpd.sgi.com's message of 19 Nov 92 06: 31:49 GMT
- Message-ID: <BOB.92Nov23235837@volitans.MorningStar.Com>
- Sender: news@MorningStar.Com
- Nntp-Posting-Host: volitans.morningstar.com
- Organization: Morning Star Technologies
- References: <1992Nov16.213034.27895@eos.arc.nasa.gov>
- <BOB.92Nov18175532@volitans.MorningStar.Com>
- <shvdb6c@rhyolite.wpd.sgi.com>
- <sjsv1fo@rhyolite.wpd.sgi.com>
- Date: Tue, 24 Nov 1992 04:58:42 GMT
- Lines: 45
-
- In article <shvdb6c@rhyolite.wpd.sgi.com> vjs@rhyolite.wpd.sgi.com (Vernon Schryver) writes:
- In article <BOB.92Nov18175532@volitans.MorningStar.Com>, bob@MorningStar.Com (Bob Sutterfield) writes:
- Even with only a single FTP stream traversing the link in one
- direction, the returning TCP ACKs (even VJ-compressed) usually
- require a large PEP packet,
-
- Are you sure? I thought the magic number was 12 bytes. 12 is much
- larger than any reasonable VJ-compressed ACK.
-
- As it was explained to me by a protocol engineer at Telebit, who had
- talked to the modem engineers, packets are sent when a timer ticks,
- not when "enough" bytes arrive. If few enough bytes are awaiting
- transmission, they'll go in a micropacket (11 bytes), which doesn't
- require that the line be turned around. If more bytes than will fit
- in a micropacket are awaiting transmission, they'll be sent in a large
- packet (340 bytes), which requires a longer line turnaround process.
- I even tried using their magical undocumented micro packet disabler
- switch (AT~D1S500=12), which should instruct the modem to always use
- small (43 bytes) packets, into which an ACK is more likely to fit, but
- that only decreased throughput and increased latency.
-
- Performance (both throughput and latency) degrades under
- increased offered load much more catastrophically with PEP than
- with V.32bis.
-
- Are you sure? From my experience, PEP does best when loaded. It
- does worst with light traffic, such as interactive stuff.
-
- I was thinking of simultaneous FTPs in both directions. Both
- throughputs suffer more than either of them does over a V.32bis
- carrier.
-
- In article <sjsv1fo@rhyolite.wpd.sgi.com> vjs@rhyolite.wpd.sgi.com (Vernon Schryver) writes:
- These kinds of latencies happen in other circumstatnces. For
- example, take a pair of Cisco routers and a 56K line. Run a TCP
- based file transfer over the link between a pair of computers with
- 60K TCP windows. At the same time, try to do some interactive
- stuff.
-
- I think cisco implements TOS queueing, but to get any benefit, you
- need to run with small MTUs (thus sacrificing some throughput) so that
- your high-priority interactive packets can squeeze ahead of the batch
- stuff that's waiting to go. If you have large MTUs, you won't be able
- to get an interactive packet on the wire until the next moby FTP
- packet is flying.
-