Subject: Re: very strange ethernet problem
To: None <port-sparc@netbsd.org>
From: Andy Brook <andyb@chp.co.uk>
List: port-sparc
Date: 11/01/2000 09:37:51
I am wondering if this is somthing to do with the Duplex status of the=20=
devices? =20=


Breif description of Duplex:
Full Duplex means that two devices can talk to each other at once,  half=20=
duplex means they must wait in turn to talk to each other one at a time=2E =20=
Generally hubs run at half duplex because it's expensive to have the clever=20=
buffering to store and forward the extra packets=2E  More clever hubs and=20=
switches can run full or half and usually auto-negotiate tho they can usually=20=
be forced one way or the other=2E

If two devices auto-neg incorrectly and one thinks it's full duplex and one=20=
thinks it's half duplex under low loads all can seem fine=2E  Start doing=20=
anything heavy and the collisions go mad because the half duplex sees the=20=
extra traffic (the machine talking at the same time as data in the other=20=
direction) as collisions=2E

I had this problem with a 3Com NIC talking to a Cisco 3524XL Switch=2E  All=20=
would be fine at low loads but try doing anything major and traffic went down=20=
to only a few hundred bytes per second=2E

I would advise checking what ALL the machines are running at and whether your=20=
hubs are clever enough to do full duplex and making sure they're all doing=20=
the same=2E  Set it manually if you can to avoid a sometimes problematic=20=
negotiation=2E  I'm not sure of the command to find this out on NetBSD being=20=
relativly new to NetBSD but your problem sounds more like a network issue=20=
than with the OS=2E

Hope you get it sorted=2E

Andy

-----Original Message-----
From: MIME :phiber@radicalmedia=2Ecom [mailto:MIME
:phiber@radicalmedia=2Ecom]
Sent: 31 October 2000 10:23
To: DON=2ELEWIS@TSC=2ETDK=2ECOM
Cc: TOD@DCA=2ENET; JEREMY@BAYMOO=2EORG; PORT-SUN3@NETBSD=2EORG; Andy Brook
Subject: Re: very strange ethernet problem



I have two hubs in two different rooms=2E  The "A" hub has a half-dozen=20=
different
machines, including the sun3=2E  I just did a test, and enabled SQE on the=20=
sun3's
transceiver=2E  Hub "B" is disconnected from hub "A"=2E  The throughput of file
transfers between the sun3 and other machines on the same hub has dramatically
increased=2E  But between any other two machines on the same hub, it's horrible=2E
Tons of collisions=2E  If I disconnect the sun3 from hub "A", everything is
normal again=2E

Now, I connect hub "B" to hub "A", without the sun3=2E  Hub "B" has the sun4,=20=
and
a freebsd/intel machine=2E  While hub "B" is connected to hub "A", there are
tons of collisions=2E  Whether between machines on hub "A", or machines on hub
"B" to "A"=2E  Reconnecting the sun3 to hub "A" just makes things worse=2E
If I disconnect the sun3, and hub "B" from hub "A", all the machines on hub=20=
"A"
are happy=2E  Throughput is wonderful=2E  Introducing either the sun3 OR hub "B"
into hub "A" trashes the throughput on hub "A"=2E  I've tested this extensively=2E
Even if the sun4 isn't connected to hub "B", connecting "B" to "A" trashes
"A"=2E  Doesn't make much sense, does it=2E=2E=2E

Thanks,
-Mark