Thanks to:
Dr. Tom Blinn <tpb_at_doctor.zk3.dec.com>
alan_at_nabeth.cxo.dec.com
Basically, clustering two 3000-500's is _possible_... They don't support
any sort of memory channel, so any cluster we build would not work under
Tru64 5.0. Also, eventhough single-ended SCSI would work on a shared
bus, it is sort of a hack. Differential SCSI is recommended for this
purpose. Further, building a cluster of such old/obsolete hardware
would cost more in time and money than the hardware itself is worth.
Dr. Tom Blinn recommends two DS10's or DS20's, differential SCSI and
memory channel for a cheap, highly available cluster.
Thanks for the help!
Responses are listed below:
From: "Dr. Tom Blinn, 603-884-0646" <tpb_at_doctor.zk3.dec.com>
> We have a DEC 3000-500 here that does medium-load web service and mail
> service. This system does its job well, so the hardware itself doesn't
> really need to be upgraded for speed, etc (at least, not yet :) It currently
> runs Tru64 4.0F.
>
> What *is* needed is a reliable system. Another DEC 3000-500 has become
> available, so we are looking into clustering the two. Basically, what we
> would like to have is this: if one server crashes, the other takes over to
> receive mail and handle web requests, etc. Even better would be some sort
> of load balancing when both systems are up, but it is not a requirement.
> After looking over the documentation, I believe we need TruCluster Available
> Server.
>
> After looking over the hardware requirements, I have a couple questions:
> 1.) Is the onboard SCSI controller on a 3000-500 a PMAZC ? I cannot find our
> 3000-500 manuals because we are in the process of moving stuff around.
> Also, Compaq seems to have forgotten to copy a lot of pages from
> www.workstation.digital.com to the new site.
No, it's older. And it's not differential. To do ASE, you need to have
a differential SCSI controller unless you can manage to keep the cables
REALLY short. Good luck doing that.
> 2.) If the onboard controller is not a PMAZC, can anyone recommend a good
> place to purchase one? I was quoted $700 for one -- the systems themselves
> aren't worth much more than that.
Good luck. It's OLD and OUTMODED and if you really need it, you're
going to have to pay a premium to get it. Plus, if you really want
to run ASE, it's not what you need, what you need is differential,
and for a TURBOchannel system, and you'll need two of them. Plus,
you'll need a suitable shared bus storage enclosure, possibly with
a RAID controller unless you can find FWD (fast wide differential)
disks. The usual choice is an HSZ40, and that will probably cost
a LOT more than the FWD controllers.
> 3.) Do the disks HAVE to be RZxx disks? The disks are currently Seagate
> ST39173N (9.1G) and ST32550N (2.1G). I would like to use these drives, if
> possible.
The disks can be just about anything, provided you meet the requirements
for shared storage. But the ONLY things that are SUPPORTED are exactly
what is listed in the relevant configuration guides (which for V4.0F
will be the TCR V1.6 guides).
> 4.) Finally, is it really worth it considering the age of these systems?
> I don't know much about the overhead of TruCluster.
The overhead of TruCluster isn't the issue. It's the cost of all of the
bits and pieces you'll need to make it work, plus the fact that ASE
style clustering is a dead end, in the next major release (V5.0 of the
Tru64 UNIX product with the corresponding TruCluster software), you've
GOT to use systems that support memory channel, and there is not now and
never will be a memory channel implementation for the DEC 3000 series.
> Thanks for any help. I will summarize to the list.
>
> --Eric
If you were asking my advice (and I realize you're not), I'd tell you to
go get a couple of uniprocessor DS20 systems, or maybe a couple of DS10s
if you're strapped for cash and floor space, and get memory channel and
the related pieces to have differential SCSI and SCSI attached RAID, and
you can build a high-performance highly-available cluster that will work
a WHOLE lot better than what you've got now and be supported well into
the future. By the time you find all the bits and pieces to try to get
TCR 1.6 running in ASE mode on the DEC 3000 systems, you will have spent
more in combined time and cash than what it will cost you to do it right
using current hardware.
Tom
-=-=-=-=-=-=-=-=-=-=-=-=
>From alan_at_nabeth.cxo.dec.com Mon Aug 16 12:05:32 1999
I believe the built-in adapters on the Model 500 were
slow SCSI and equivalent of the PMAZB.
Only RZ disks were ever qualified. Without being
qualified there's no way we can guarantee that it
will work.
-=-=-=-=-=-=-=-=-=-=-=-=
>From tpb_at_doctor.zk3.dec.com Mon Aug 16 12:07:13 1999
> On Thu, 12 Aug 1999, Dr. Tom Blinn, 603-884-0646 wrote:
>
> >No, it's older. And it's not differential. To do ASE, you need to have
> >a differential SCSI controller unless you can manage to keep the cables
> >REALLY short. Good luck doing that.
>
> The reason I began considering this is because in the Hardware Configuration
> manual for TCR, it has an example of something very similar to what I would
> like to do -- two 3000-500's, SE SCSI and a BA350 enclosure. Is this wrong?
> Or is it just good as an example (ie, not good in real life) ?
I'd have to dig out the manual and double check. The BA350 can be set
up to sit between two SCSI busses, and if the cables are short enough
you don't need to have differential SCSI. But it's sort of a hack. I
can't say that I've ever heard of anyone around here doing it that way
but it just might work..
Received on Tue Aug 17 1999 - 23:41:14 NZST