home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!olivea!decwrl!elroy.jpl.nasa.gov!ncar!mrbean.scd.ucar.edu!hyder
- From: hyder@mrbean.scd.ucar.edu (Paul Hyder)
- Newsgroups: comp.dcom.lans.ethernet
- Subject: Re: Early and Late Collision
- Message-ID: <1992Jul31.155839.15676@ncar.ucar.edu>
- Date: 31 Jul 92 15:58:39 GMT
- References: <1992Jul30.003641.22901@news.iastate.edu> <1992Jul30.224300.14298@ncar.ucar.edu> <nvhgb2k@rhyolite.wpd.sgi.com>
- Sender: news@ncar.ucar.edu (USENET Maintenance)
- Organization: Scientific Computing Divison/NCAR Boulder, CO
- Lines: 42
-
- >> > On many hosts it is impossible to capture long collisions, only a few
- >> > hosts even bother to tell you about them. You need something like
- >> > a LANalyzer to capture the packets. It is important to know how far
- >> > into the packet the collision was, close values point to cable length
- >> > and longer ones other problems.
- >>
- >>
- >> How can you capture a late collision? What will you see but the jumble
- >> of two simultaneous transmissions? The bad guy will be the one whose
- >> first bit is in trashed by the later bits of the innocent guy. The
- >> easily capturable MAC header will be that of the innocent guy.
- >>
- >>
- >> Does the LANalyzer or anything else have enough electronics to try to
- >> peel some of the bits apart by assuming the two clocks are different?
- >> Or does it compare the signal at several different points on the
- >> network, and thereby figure out the bad guy's bits?
-
- <sigh, another broken promise. ... just figure I lied.>
-
- Slight difference in the meaning of "capture" is the key here. All
- one can easily do is snag a copy of whatever garbled mess was on
- the net. I don't know of any magic to strip it logically back apart.
-
- Late collisions occur far enough into a transmission that you can
- often extract the address info and sometimes get clear to the protocol
- headers. As is indicated in the query, you can't tell who or what is
- doing the stomping but you can tell who had one of the packets that
- will be dropped >AND< how far into the packet the collision occured.
- In crisis cases samples taken at various points on the net might give
- a set of hosts to start checking. [Moving around has never worked for
- me but then I've also never had the problem end up being a cable that
- was too long.]
-
- Most enet chipsets discard bad packets (CRC, late collision, etc) so
- you can't actually look at them on a host via utilities like tcpdump.
- Custom HW in lan monitors captures most bad packets, sometimes all
- you get is jumbled bits that can only loosely be called a packet.
- Much of the time there are helpful readable bits at the beginning.
-
- Paul Hyder
- NCAR
-