Re: [9fans] Alpha success (w/ clock.c bugfix and ether2114x.c update)

[Ronald G Minnich <rminnich@lanl.gov>]

On Tue, 23 Oct 2001 jmk@plan9.bell-labs.com wrote:

> So, I asked if anything had changed with the Procurve switch and the
> answer was no, well, actually we upgraded the firmware in all of them by
> rotating them out with a spare. I tried the other DE-500 based system
> and found it wouldn't talk to the Procurve either. Somehow the upgrade
> had been done without taking the systems down so fortunately the
> fileserver was still working (no idea how, but miracles sometimes happen).
> I had to replace the ether card in the fileserver as, sure enough, after
> a reboot it too couldn't talk to the Procurve.


Ah yes, the infamous "All the good people left DEC so we can't design
ethernet cards anymore" problem with Alphas. We have a few hundred of
these machines here. DEC never seems to have gotten auto-negotiation right
in the last few years, and once you figure it out and fix it, it doesn't
work anyway on the next little rev of the hardware. The problem always
seems to boil down to very small timing problems such that the card comes
up half and the switch full, or the switch half and card full, or both
full, but the PHY is messed up so they can't talk anyway (the last time we
fixed this). The usual way we fix it is to tell the switch to skip
autonegotiation and wire the port to 100 Full. I mean, really, it's 2001,
so 10 mbit half is unlikely anyway. Hardwiring switch ports works until it
doesn't, and then you tweak the driver. Which will work until the next
broken Alpha hardware comes in -- which won't be happening for long, so
this will be less of a problem in future.

I've never been able to figure out how these guys did so well with their
ethernet chips years ago and then did nothing but screw up the last five
years. I'm convinced all the good guys quit and went to other companies --
that would explain a lot.

ron