I continue to try and find out why the performance of all our Alphas went
to hell when I "upgraded" from 2.0 to 3.2. Today's suspect -- the
tulip ethernet interface.
All the tulip-based machines (about 10) are consistently worse than
the lance-based machines (about 20) in regards to collisions. Most
of the tulip cards show about 10% to 25% collision rates. Most of
the lance cards show 1% to 8%. One machine is particularly bad.
The two previous times I checked, it had 90% and 60% collision rates.
I rebooted the machine this morning. Now it shows
Name Mtu Network Address Ipkts Ierrs Opkts Oerrs Coll
tu0 1500 <Link> 08:00:2b:e4:bb:d8 1273834 0 348590 2 544050
for a whopping 156% collision rate! (collisions/output_pkts) I've
gone by the rule that anything above 10% is unacceptable. In my
experience with HP and Sun machines, I rarely see anything above 2%.
The Alphas seem to average about 4% (ignoring outliers). None of
this makes me feel good.
The boot message shows
Dec 20 09:45:26 basie vmunix: tu0: DECchip 21040-AA: Revision: 2.3
Dec 20 09:45:26 basie vmunix: tu0 at pci0
Dec 20 09:45:26 basie vmunix: tu0: DEC TULIP Ethernet Interface, hardware address: 08-00-2B-E4-BB-D8
Dec 20 09:45:26 basie vmunix: tu0: selecting 10BaseT port
There is a strong correlation on our tulip boxes showing that older
machines have worse collision rates. The above machine was one of
our first alphastation purchases.
All machines are running 3.2B, the upgrade to 3.2C is in progress.
All wiring is 10baseT using commercially made cables (AMP). The
infrastructure wiring is all Cat-5. All systems go to Cabletron hubs.
Is there something fundamentally wrong with tulip boards? HELP!
Jim Wright Keck Center for Integrative Neuroscience
jwright_at_phy.ucsf.edu Box 0444, Room HSE-802
voice 415-502-4874 513 Parnassus Ave
fax 415-502-4848 UCSF, San Francisco, CA 94143-0444
Received on Thu Dec 21 1995 - 10:56:43 NZDT