[geeks] Sun Fire V210: A New Low in SPARC Servers?

[Jonathan C. Patschke]

I have to admit to having not purchased many new Suns in the last few
years.  The newest machines I have at home is a Netra X1.  Still, aside
from its relatively poor disk performance, it's a nice system.  The
Netra T1s I have are just fabulous--an excellent example of a well-built
piece of computing hardware from Sun's dot-com heyday.

$ork just bought five Sun Fire V210 systems to act as headless crunch
boxes for solving semiconductor layout constraints.  I was -really-
excited to get my hands on some US-IIIi server-class hardware, after
than our barely-better-than-a-PC Blade 1500 systems.

The V210 is a well-built machine.  It has the nicest rack-rail system
I've seen in a very long time (although the cable management arm is
utter junk).  The machine it easy to work in.  The front access panel is
thoughtfully designed so that it will flip down with enough clearance
for the operator to remove and install disks without pulling the system
out of the rack.  The fan tray is a little hokey, but if four 0.5A,
-13000rpm- fans can't keep the system cool[0], nothing short of water-
cooling will.

And that's about where the fun stops.  I knew I was in for a wild ride
when I unpacked the server and saw a sticker alerting me to the fact
that maximum POST/OBP diagnostics are enabled on -every boot- by default
to ensure maximum system serviceability or some garbage.

We ordered systems with Solars 8, Solaris 9, and Solaris 10, in hopes
that we could get the newest media kits for each.  We got Solaris 8 2/04
(the latest/last, as far as I know), Solaris 9 8/04 (a year behind the
latest release), and Solaris 10 3/05.  I can understand Sun not having a
fast enough shipping pipeline to get 10 1/06 into an order that shipped
a few days after its release, but not including the 8/05 release of
Solaris 9 is puzzling to say the least.

ALOM (the replacement for LOM in the Netra T1/X1) is fully network-
capable.  You can query the MAC address of the ALOM subsystem from the
serial port, pop that into your DHCP server, run the "setupsc" script in
ALOM to configure remote access with an address obtained via DHCP,
restart ALOM, and have all the benefits of a console connection from
over the network.

At least, that's how it's supposed to work.  On all the machines I've
tried so far, running "setupsc" followed by "resetsc" has resulted in a
fully dead ALOM that hangs while loading itself from flash.  Sun came
out today to replace the system board in one of the machines, but I'm
apprehensive about trying this feature again without at least seeing if
the ALOM firmware is as new as possible.  The service engineer pointed
me at a jumper on the system board that's supposed to (but apparently
doesn't) reset ALOM and the system NVRAM to recover from this condition.

So, I've been debugging strange JumpStart problems on the machines that
ALOM hasn't killed yet.  We just migrated to JumpStart over DHCP.  I
could JumpStart anything I wanted over RARP + RPC, and now I can
jumpstart things over DHCP, but not the V210s:

   Sun Fire V210, No Keyboard
   Copyright 1998-2004 Sun Microsystems, Inc.  All rights reserved.
   OpenBoot 4.16.2, 512 MB memory installed, Serial #xxyyxxyy.
   Ethernet address 0:3:ba:xx:yy:xx, Host ID: xxyyxxyy.



   Rebooting with command: boot net:dhcp - install
   Boot device: /pci at 1f,700000/network at 2:dhcp  File and args: - install
   100 Mbps FDX Link up
   4000 100 Mbps FDX Link up
   SunOS Release 5.8 Version Generic_108528-29 64-bit
   Copyright 1983-2003 Sun Microsystems, Inc.  All rights reserved.
   Hardware watchdog enabled

And there it hangs.  I can't even break back into OBP; I have to get
ALOM to power-cycle the system.

Also, apparently Sun's web-based ticket-opening system recognizes
neither the chassis serial number nor the OBP-reported serial number nor
the system hostid.  The only way I could open a ticket was by speaking
with three fellows with extremely thick accents over phone support who
still managed to misdocument[1] the problem.

If I wanted unreliable junk and lots of service-hassle, I could buy it
at a much lower price from a certain vendor a mile up interstate 35.



[0] and audible with a peak noise output of 51dB per fan.
[1] "I am to be correct in understanding that you have problem with
      E-LUM not fully analyzing?"  "Sure, fine, close-enough, the CE will
      just call me when he gets the ticked."
-- 
Jonathan Patschke    )   "A man who never dreams goes slowly mad."
Elgin, TX           (      --Thomas Dolby, "Valley of the Mind's Eye"