home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!haven.umd.edu!darwin.sura.net!mips!swrinde!elroy.jpl.nasa.gov!usc!noiro.acs.uci.edu!gordius!gordius!bruce
- From: bruce@gordian.com (Bruce Hafford)
- Newsgroups: comp.sys.mips
- Subject: R4000 & Mips OS 5.0b = horrid TCP behavior??
- Keywords: R4000,TCP/IP
- Message-ID: <1992Aug12.223059.28974@gordian.com>
- Date: 12 Aug 92 22:30:59 GMT
- Sender: news@gordian.com
- Organization: Gordian
- Lines: 40
-
- Well, we finally got our new Mips R4000 station booted and
- happy. Except for having the wrong Xwindows version, but that's
- another story. Anyway, we're seeing some gross TCP/IP behavior
- and are wondering if anyone else has noticed this.
-
- There seem to be 3 (possibly interrelated) problems:
-
- 1) WAY too many plain TCP ack packets. If the R4000 is generating
- data to a Telnet user, for example, it will follow a slew of data packets
- with a slew of ACKs, even though the user did not send anything
- that needs acking. These acks number 10-25 at a time, spaced at
- about 700usec - 1.5 msec apart. disgusting!
-
- 2) Unbuffered data output. On the above user, the R4000 generates data
- packets (~50 bytes each) every ~1msec. Slower TCP clients get
- completely swamped, and quickly fall behind. Buffering the output
- is sorely needed, if only for 2-5 msec. Or at least dynamic modification
- of the output timer intervals?
-
- Between items 1 & 2, the packet ratio ends up with the R4000 sending
- about 6-7 times as many packets as the listener. OS 4.52 on a Magnum 3xxx
- end up about 1.5 - 2 times as many, if that.
-
- 3) Bogus retransmissions? We have a couple products here that do
- not keep up with the R4000's TCP rate, and thus fall behind and drop
- packets. The R4000 seems to handle retransmissions ok sometimes,
- but several times the session has completely stopped dead because
- the 4000 did NOT retransmit the correct packet. The other TCP node
- continued to ack up to that dropped packet, but the 4000 never
- resends it. Needless to say, incorrect as far as we can tell.
-
- If these things are tunable in the OS, that's fine, but having the
- default be "too many packets, dropped retransmits" seems wrong-ish.
-
- So - anyone else out there using the R4000 and new 5.0b Mips OS?
- Have you seen anything like this? email or post - I'll summarize
- if people are interested. And if any Mips engineers happen to
- run across this, feel free to contact us!
-
- bruce@gordian.com
-