>>>>> "gw" == Greg A Woods; Planix, Inc <woods%planix.ca@localhost> writes:
gw> I will further concur that in general an AlphaServer is still
gw> a very decent platform for NetBSD,
fine, but in addition to the known problems with IDE on some models
please also remember that threaded programs do not work.
Also, the last straw for me was that ddb and gdb do not work reliably
any more, to do simple things like backtraces on kernel core dumps, so
when your kernel does panic, it's much more difficult for what
interested NetBSD developers there are to help you with the problem
than it would be on a different platform.
gw> The problem on the ES40 may, or may not, be directly related
gw> to NetBSD drivers
It doesn't really matter to someone choosing a platform whether the
problem is with the chip or the drivers. It just matters that there's
a problem, unless you can avoid the problem device.
In my DS10, IIRC, I found that neither the built-in IDE nor the add-on
Promise card (which works fine in other machines) worked. But this
may be that the Promise card needs some registers set up by its 8086
firmware. so, there's another ``data point'' albeit as hazy as
anything.
What would also matter is if the problem were in NetBSD (not chip),
but NOT in the IDE driver. PCI in the distant past has been a bit
weird on the alpha:
http://mail-index.netbsd.org/port-alpha/1999/07/11/0009.html
http://mail-index.netbsd.org/port-alpha/2001/02/26/0003.html
The problem was never really isolated, but people stopped complaining
about it:
http://www.netbsd.org/cgi-bin/query-pr-single.pl?number=7826
It sounds like a pretty obvious problem, so i think people would
complain about it if it were still around. OTOH it was amazing how
few people complained about it back when it existed---for a problem
that stuck around for >1.5 years, no one complained loud enough to
stop this poor guy from sinking >$30k into the platform, and then
running it, with known severe problems, at 10Mbit/s. That's, um,
_bad_. I don't have the problem, but that could be because I don't
use IDE any more, or because the few devices I do use (tlp, QLogic
ISP) are fixed but in general the original problem is not fixed.
I intentionally only use cards in alphas that almost everyone else is
also using. I've also had problems with a quad-tulip LSI Logic card
that paniced SRM on boot, before NetBSD got control, and a
USB+Firewire card that paniced something, I forget if it was NetBSD or
SRM. I have a quad-tulip Adaptec card that panics a DS10, but does
not panic an Avanti. I tested the Adaptec card in a friend's avanti
and thought it was good-for-Alpha, then bought three more of them and
found it wasn't, and had to go buy an avanti. That kind of pissed me
off.
In the Avanti, the quad-tulip card has only about 10kpps forwarding
performance (it has trouble even with 100Mbit/s of big packets, and
can be DDoSed by a syn flood over DSL). I also get transmit and
receive under- and overrun errors from tlp(4) so that it degrades to
larger and larger buffers and then finally to ``store-and-forward
mode'', which might be normal, or might indicate some kind of PCI
arbitration problem (the card can't DMA with low enough latency?).
I replaced the alpha with FreeBSD/sparc64 and an em/wm card, and now
I'm getting panics, I think, whenever the card receives a jumbo frame,
not sure yet.
Overall I've found alphas are fun machines that impress others and are
somewhat nice to be around, but I've grown somewhat suspicious of them
in general, and specifically thoroughly tired of dealing with
extremely basic problems that no one else around me is having.
Attachment:
pgpwX6eSx0tLP.pgp
Description: PGP signature