home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!math.fu-berlin.de!mailgzrz.TU-Berlin.DE!wncs.zrz.TU-Berlin.DE!kwia4000
- From: kwia4000@wncs.zrz.TU-Berlin.DE (Manfred Kwiatkowski)
- Newsgroups: comp.protocols.tcp-ip
- Subject: Re: Multiport repeaters and collision propagation
- Date: 27 Dec 1992 03:14:12 GMT
- Organization: Tech. Univ. of Berlin, Central Computing Facility
- Lines: 44
- Distribution: world
- Message-ID: <1992Dec27.041413@wncs.zrz.TU-Berlin.DE>
- References: <ric.724092787@updike> <1992Dec14.193852.28204@alias.com>
- Reply-To: kwiatkowski@zrz.tu-berlin.de
- NNTP-Posting-Host: wncs.zrz.tu-berlin.de
-
- In article <1992Dec14.193852.28204@alias.com>, chk@alias.com (C. Harald Koch) writes:
- |>
- |> What I'm seeing is strange; I can blast data over a TCP/IP connection at
- |> over 800Kb/s from host A to host B, but throughput is around 20Kb/s in the
- |> reverse direction. Blasting UDP data across the network shows extremely high
- |> packet losses in one directory.
- |>
- |> There are two multi-port repeaters between the hosts; all cabling is
- |> thin-Ethernet. All of my test hosts are the same (SGI Indigos with Allied
- |> Telesis Micro-transceivers, all running the same OS revision).
- |>
- |> Hosts that are both on the same local thin-net cable (i.e. not talking
- |> through a repeater) can talk to each other at high-speed, bi-directionally,
- |> without problems. Also, not all repeaters show these symptoms.
- |>
- |> I've been monitoring packets on the cables lately to try to figure out
- |> what's wrong. If I run monitors on machines 'next to' A and B, I see many
- |> packets transmitted by machine B, but that never appear on machine A's
- |> cable. The only conclusion I've reached so far is that for some reason, the
- |> repeater is dropping packets. (This is supposed to be impossible, due to
- |> collision propogation and Ethernet's exponential backoff). Incidentally,
- |> netstat on the transmitter shows no output errors and no dropped packets.
-
- Two common causes for 'lost packets' on Thinnet-MPRs are:
-
- a) exceeded segment-length limit.
-
- Due to their task, repeaters have to be most 'picky' concerning
- signal levels (receive collision detection scheme) and usually
- operate flakely first in an out-of-specs environment where local
- end-nodes may still communicate. Your observed unsymmetric behavior
- may be due to packet-length dependencies (data >> ACK)
-
- b) grounding problems.
-
- MPRs with builtin-terminators usually ground the cable. If there
- is a second ground somewhere (e.g another MPR!!) the potential
- shift may cause peculiar latch-up effects ( watch the collision
- LED)
-
- Hope this helps...
-
- --
- Manfred Kwiatkowski kwiatkowski@zrz.tu-berlin.de
-