[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Requalification of Alpha for etch

On 11 Oct 2005, at 1:33 pm, Thimo Neubauer wrote:

Ok, that would perfectly match goedel (also dual, also 4G) :-) One
question though: the Symbios Logic 53c895 SCSI-controller in goedel
acts weird in SMP mode (tried some 2.4 and 2.6 kernels) which is why
the machine runs in single CPU mode all the time. Does it work for

We don't currently run Linux on any of the larger servers, so I haven't had any experience of that yet. Certainly haven't had any trouble on DS10Ls, which are the largest machine I have yet tried.

Judging by the SRM show config output, this machine does not use that SCSI controller, but has dual QLogic controllers:

P00>>>show config

SRM Console:    V6.0-8
PALcode:        OpenVMS PALcode V1.92-73, Tru64 UNIX PALcode V1.87-69

CPU 0           Alpha EV67 pass 2.6 500 MHz     SROM Revision: V1.82
                Bcache size: 4 MB

CPU 1           Alpha EV67 pass 2.6 500 MHz     SROM Revision: V1.82
                Bcache size: 4 MB

Core Logic
Cchip           DECchip 21272-CA Rev 2.1
Dchip           DECchip 21272-DA Rev 2.0
Pchip 0         DECchip 21272-EA Rev 2.2
Pchip 1         DECchip 21272-EA Rev 2.2

TIG             Rev 4.14
Arbiter         Rev 2.10 (0x1)


Array #       Size     Base Addr
-------    ----------  ---------
   0         1024 MB    000000000
   1         1024 MB    040000000

Total Bad Pages = 0
Total Good Memory = 2048 MBytes

PCI Hose 00
     Bus 00  Slot 05/0: Cypress 82C693
Bridge to Bus 1, ISA
     Bus 00  Slot 05/1: Cypress 82C693 IDE
     Bus 00  Slot 05/2: Cypress 82C693 IDE
     Bus 00  Slot 05/3: Cypress 82C693 USB

     Bus 00  Slot 07: DAPBA-UA ATM155 UTP

     Bus 00  Slot 08: QLogic ISP10x0
                                   pkb0.          SCSI Bus ID 7
                                   dkb0.           RZ1CF-CF
                                   dkb100.         RZ1CF-CF
dkb600. COMPAQ BA03611C9B mkb300. COMPAQ SDX-300C
     Bus 00  Slot 09: DEC PCI MC
                                   mcb0.          Rev: 22, mcb0

PCI Hose 01
     Bus 00  Slot 07: DEC PCI MC
                                   mca0.          Rev: 22, mca0
     Bus 00  Slot 08: KGPSA-B
pga0. WWN 2000-0000-c921-4
                                   dga11.1001.0.8.1       HSG80
                                   dga11.1002.0.8.1       HSG80
                                   dga11.1003.0.8.1       HSG80
                                   dga11.1004.0.8.1       HSG80
     Bus 00  Slot 09: QLogic ISP10x0
                                   pka0.          SCSI Bus ID 7
                                   dka500.         RRD47

The KGPSA and Memory Channel controllers will not be there when Debian gets the machine (or at least won't be connected to anything - we don't want to keep the HSG80 arrays going, the maintenance is too expensive -- I'd probably use software RAID1 on internal disks instead)

Or do you have a different controller?

OK, I've double checked with the boss. We're decommissioning an 8- node DS20E Tru64 cluster this week, and he's happy for us to have one of the machines. This will be a dual-CPU 500 MHz EV67 machine with 2GB RAM (not quite what I was hoping for, but hopefully good enough). I hopefully can grab some of the RAM out of one of the others to beef up the new Debian machine.

Within a few months he says we can replace it with a quad CPU ES45 once those clusters are decommissioned, which will be 1GHz CPUs at least, and depending on which machine we get, could have anything from 4-16 GB memory.

Ok, so I guess that your offer fits all conditions so I'd say that we
have a new local buildd admin :-) Any objection?

Nope. Got boss approval, and there are three DD's in our systems team (myself, Dave Holland and Simon Kelley) who will be happy to do any console stuff that might be required.

Please speak

The next steps for the machine to become a buildd would be:

 * send "debian-admin" a mail that you're preparing a new machine as a
   buildd so that we can qualify for etch. Ask, who of the admins
   would like to get a superuser login.

OK, I'll do that today.

 * install a fresh Sarge on the box, best with no real user so there
   won't be UID clashes

Fair enough.

 * on goedel and escher there are "adminloginR"-accounts with uid 0 so
   that the Debian Admins can do whatever they want and you can keep
   being "root". If they tell you who will care for the start (I guess
   that'll be Joey) send him a GPG-crypted password

 * wait for a while (can be weeks)

:-) It'll be at least a week for us to install the machine anyway; the transition of the current operational cluster to its new hardware needs to have completed successfully this week (it's scheduled for Wednesday, and they'll want to test for a couple of weeks, this being a 24/7 DNA sequencing production cluster) and then I'll need to physically move the machine since the current cluster's rack will be being dismantled to make room for something else, and it needs to be moved to a DMZ rack anyway.

You can of course choose a name but I'd recommend "bach". First, it
definitely fits the original naming scheme


of "classical composers (including baroque, romantics and
impressionists) of choir music" and secondly the Alpha machines would
form the "Goedel, Escher, Bach"-triple[1] :-)

Sounds good to me. I approve - singing in choirs is one of my other hobbies...


Reply to: