rtp1@quads.uchicago.edu (raymond thomas pierrehumbert) (07/10/90)
After weeks and weeks of waiting, the gold-plated $7500 Maxtor
disk for my DN10000 (you know, the kind that costs $2200 with
a 2 year warrantee if you buy it as a SCSI disk for a Sun) finally
arrived. The technician installed it and invol'd it, but he knows
no Unix at all, and couldn't figure out how to mount it in my
BSD4.3-only environment. He called somebody back at the office
and still couldn't figure it out. Jeez, this is ridiculous.
Can somebody tell me what I need to do? I imagine that
I need to create some kind of file in the /dev directory, maybe
something like wn1a (controller 1,first device). How do I do
this? The man pages on mkdev are unhelpful. Then how do I mount
it? The man pages on mount seem to suggest that it is only for
removable file systems, like tape, or remote file systems. Is
it true, or by "removable" do they also include internal disks
that cannot be physically removed? Anyway, if /etc/mount is
what I need, what arguments does it take to mount a winchester
disk? Please help, I'm starving for disk space.
.krowitz%richter@UMIX.CC.UMICH.EDU (David Krowitz) (07/10/90)
Walt Webber, are you there? HP/Apollo division should be paying
me commissions for user-support!
OK, this comes from the Update to Domain Series 2500 Owner's Guide,
order number 017934-A00 ...
Use the regular "invol" program to format the disk. If you are adding
a disk to a system which is already up and running, you can do this
online using /etc/invol (/com/invol is a link to /etc/invol for you
Aegis users). Use option 1 -b to initialize the disk with BSD-style
ACL's. If you want to boot from this disk, then use option 8 to
create an OS paging file after you do option 1. Invol prompts for
the disk to use in the form "w<controller>:<unit>" ie. to init the
second disk on the first controller, you would use "w0:1" -- to init
the first disk on the second controller (ie. the 3rd disk in the system
as each controller can handle two disks), you would use "w1:0". Be aware
that the "controller" and "unit" have different meanings for ESDI (ie.
DN3xxx/4xxx and DN10000) disks and SCSI disks. With the ESDI disks,
"controller" is what you think it means -- the number of the disk
controller board, ie. 0 or 1 -- and "unit" is the number of the disk
attached to a particular controller, ie. 0 or 1 again. With SCSI disks,
"unit" is always 0, and "controller" is the number of the SCSI target
ID which is set on the disk (i.e 0 to 6 -- 7 is reserved for the CPU's
target ID).
Once you have involed the disk, you can create the the Unix device
file with the command: /etc/mkdsk W <controller> <unit> <logical volume>
where <controller> and <unit> are the same as they were in the "invol"
command, and <logical volume> is the number of the logical volume you
created when you initialized the disk ("invol" will allow you to partition
a disk into 1 or more logical volumes). For example, to create the Unix
device file for the 1st logical volume of the second disk (ie. unit 1) on
the first controller (ie. controller 0) you would use the command:
/etc/mkdsk W 0 1 1 To create the device file for the 2nd logical volume
of the same disk, you would use /etc/mkdsk W 0 1 2.
Once you have created the device file, you make an emptry directory for
a mount point, just as in a native Unix system, using /bin/mkdir (ie.
/bin/mkdir /moint_point_for_new_disk) and then use the regular Unix
"mount" command to mount the disk using the device file you just created
(ie. /etc/mount /dev/whatever_mkdsk_reported_as_the_device_name /mount_point).
-- David Krowitz
krowitz@richter.mit.edu (18.83.0.109)
krowitz%richter.mit.edu@eddie.mit.edu
krowitz%richter.mit.edu@mitvma.bitnet
(in order of decreasing preference)rtp1@midway.uchicago.edu (raymond thomas pierrehumbert) (07/11/90)
>use /etc/mkdsk ...
That's fine, but I'm running 10.1p, not 10.2p, because (a) I don't
have enough disk space to load 10.2p until I get the bloody second
disk mounted, (b) People have reported so many problems with 10.2p
I'm afraid to load it until I can afford a lot of hassle and downtime,
which may be never, (c) I need to get the version of NFS that runs
with 10.2p. I know this is going to be a hassle, if it is anything
like what I went through to get Fortran 10.7p, and I am not very
confident of my sales reps assurances that it really works. The
800 number said something like "it works but you can't mount
directories, only files." Ha.
When I installed BSD_medium at 10.1p, /com/mountvol was not
loaded, so it looks like this will have to get installed on my
machine somehow. The tech was not able to get it off my 10.1p
tape, so he had to go back to the office and make a tape there
(I'm still waiting).
This shows how HP/Apollo customer support as we now know it
is not only making irate customers, but is also bad for business.
It's not the tech's fault-- it's the fault of the information
distribution and training system. He would up spending 5 hours
at my site (which must have been 5 EXPENSIVE non-billable hours),
and only at the end was able to find out that at 10.1 you can't
do it in UNIX. Then he has to make another visit with a tape--
so I'm on Internet, why isn't there some place he could have
ftp'd it from? Go figure.
As to the "why buy Apollo" thread, given my experiences with
customer support and third party hardware (or lack of it) I think
you'd have to be crazy to buy any of the Apollos below the DN10000,
because Sun does the same thing so much better and cheaper (unless
you have an existing investment in Apollo hardware or software
to protect.) You are also insane to buy a 10000 if you want to
support a lot of users running small jobs; a cluster of Sparcstations
does this much better. There is some marginal argument for buying
a DN10k for floating point crunching, which is my application. Even
there, the argument is no longer so clear to me; it's very code
dependence. I think on most of my code, I'm getting no more than
about 2 megaflops, which is about the same as a Sparc330. On a
matrix multiply (single precision) you get 5 megaflops, which is better
than twice a 330 (but still not quite at the breakeven point). If
you do a lot of saxpy, you're in luck, for a 1D saxpy goes at 10
megaflops. If all you do is innerproducts, you're even better, since
you get 28megaflops. In all, though, I would be surprised if my
average mix is better than 2-4 megaflops (1 processor).
I am so fed up, I'd sell my 10k and get something else, except
that at the high-performance end, I'm not sure the competition (
Stardent, Silicon Graphics) is all that much better. I bought the
10k because it offered an easy upgrade path (its megaflop per
kilobuck ratio gets better as you add more processors, and add on
processors are quite reasonably priced). I'm still not sure what
I'd do if I had it all to do over again (and actually I will
come September, to the tune of about $.7 million if our NSF
cloud center comes through), I'm still not sure what I'd do. I do know
that the problems I have had at my site have not been a big selling
point for 10k's elsewhere. Two sales at U. Chicago I know of fell
through partly because of information that got around about my
experiences (not that I spread it myself). I also tried to encourage
a former student now in Canada to buy a 10k for his department,
but following some honest answers about my experiences, he went
with Silicon Graphics.
My disk debacle is only the latest in a long series of tribulations.
The 10k COULD be the machine that saves Apollo, but some things really
need to change. I'm trying to be an advocate for this hunk of iron,
but HP/A is not making things easy for me.
.