Several responses came in in response to the first summary. They follow (edited
for bandwidth).
The biggest revelation in this occurred to me yesterday: A transmitter-detected
collision is probably going to be detected by most monitoring devices, especially a device not on the same piece of copper, as a misalign or a
CRC error. Thus, the collisions I saw and the misaligns and CRC errors the
networks people saw were the same errors. Realizing that, I can analyze the
problems using only netstat -i.
My conclusion is this: The collisions have dropped from 8-10% of Ipkts to
0.5-2%. That is acceptable given that there will be substantial reductions
in traffic when we add memory to the clients.
Thanks again to all who responded:
=========
----- Begin Included Message -----
From: jas@proteon.com (John A. Shriver)
Early product Sun-3/60's were running their Ethernet chips at the
wrong frequency. (They didn't put the correct shading capacitors on
the crystal.) They worked fine with other nodes on the same wire, but
caused all sorts of grief through a repeater. That is the only time
an off-frequency clock realy matters.
Well, a 10BASE-T hub is a multiport repeater. Thus, it cars about the
center frequency.
I find it hard to beleive that Sun would make so serious a gaffe as
this again, but they are a large company.
You could put an accurate counter on the Ethernet serial interface
chip. Some pin should be 20 MHz or 10 MHz. See the Ethernet spec for
the tolerance.
----- Begin Included Message -----
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:06:14 CDT