HP/DEC/Compaq

Hello there,
I used DECStations a lot when I was in college back in the days.
But I never knew how they actually worked, of course most of them have retired.
Firstly, I think we had diskless workstations sitting in the lab, and an X window
login screen, when you logged in. When you had, you waited for the X window OS to load up,
I remember the OS was called OSF/1. But the Client diskless OS was a bit like
Ultrix, so if X11R5 was the GUI server of the workstation, it should actually
reside on the workstation but if so, it is contradicting because it was diskless.
But if it wasn't there, the OSF/1 server did have to deliver the Ultrix OS to
the workstations, I remember the whole bunch of workstations freezed altogther in the lab
when the network died. So how did the whole thing work in general?
Thanks
Jack

OSF/1 was an operating system from the Open Software Foundation in Cambridge, MA. (no relation at all to free software or open source)
It was mainly used by DEC under license, on the Alpha machines. That system was initially called OSF/1, but DEC renamed it later to Digital Unix, and later under Compaq it was called Tru64.
There did exist a version of OSF/1 for the MIPS-based DECstations, but it was experimental. I don't think it was used much in business or academia. It probably was mainly used as a means to develop software for the Alpha before the hardware was ready.

Diskless workstations were an early form of thin client. The idea is that system administration and backup only need to be done for the server, and all the clients are "zero state" so require no management. In practice, they were less cost effective, because all the I/O needed to go through a very slow shared network. In order to work without a disk, the workstations needed to have code in firmware to set up networking, find a server, and load the operating system over the network. Once unix is working, it just mounts its root over NFS (or AFS or RFS) and works like a unix system with a very slow disk.

The scenario you describe, though, sounds more like the DECstations were only used as terminals, with the users logging on to their accounts on an Alpha server. That was even less cost effective, since Xterminals were available at a small fraction of the cost of a workstation, and performed the same task. But at least disk access would be fast (since it's from a session on a server, to the server's own disk). This kind of remote login from a terminal is achieved using xdm.

robespierre wrote: Diskless workstations were an early form of thin client. The idea is that system administration and backup only need to be done for the server, and all the clients are "zero state" so require no management. In practice, they were less cost effective, because all the I/O needed to go through a very slow shared network.

I'd say this view is more in line for the 1990s than the 1980s, when diskless workstations first appeared - when Apollo DN300s and Sun-2s and -3s were the new hotness. The original impetus for diskless workstations was largely the expense and performance of mass storage. A typical diskless client from that era would be running an almost-complete operating system image, not a cut-down GUI or front-end like a thin client.

What did the mass storage landscape look like, to cause this? In the early 80s you're looking at top-of-the-line hard drives of a few hundred megabytes, using 14" platters and interconnects like SMD, and weighing hundreds of pounds. An example common by the mid-80s would be the Fujitsu Eagle , a ~470MB SMD drive using 10.5" platters that would retail for $10,000. Meanwhile in 1985 Seagate announced the ST4051, a 40MB 5.25" HDD which would initially cost you almost $1,000 in OEM quantities, let alone retail, and provide such slow access times and throughput that the same workstation running diskless could often out-perform it despite having all disk I/O go over 10Mb Ethernet to a fast server with an Eagle. Larger 5.25" HDD were available, and their prices did drop rapidly in the latter half of the 80s, but it took a while for the performance equation to change.

So when buying a lab or department of workstations that were going to need a fileserver anyway, why spend $1,500-2,000 extra per machine for poor performance and little space when you could run them diskless and get a second big shared drive on the server? In 1986 such a server package (Sun-2/170, two Eagles, and a tape drive) would list for $80,000, versus $8,900 for a Sun-2/50 desktop w/o disk. The December 198 8 add-on cost for an external "shoebox" with 71MB drive suitable for the Sun-2/50 was nearly $3,000 list...


It was common in the 90s to recycle those same older workstations as Xterminals. There were even packages like "xkernel" to help do this - actually, upon Googling, it looks like there may have been a few variants. See http://www.bond.id.au/~gnb/papers/xkernel.ps.Z for a nice overview that could be adapted to any number of systems - including DECstations running Ultrix.

Anyway depending on the dates involved, using the DECstations as Xterminals was probably a decent way to avoid the expense of buying new Xterminals, and it would be straightforward to have them run a stripped down Ultrix diskless config with an X server. And as robespierre said, allow XDMCP to manage it from a (then) big, fast Alpha. And given what CPUs were being used in low-end Xterminals, you were probably getting better performance to boot.

I've just found some of those firmwares in the public domain located at hp.com
But don't know which one I should look for.

ftp://ftp.hp.com/pub/alphaserver/

lucky7456969 wrote: I've just found some of those firmwares in the public domain located at hp.com
But don't know which one I should look for.

ftp://ftp.hp.com/pub/ alphaserver /

First off, I doubt any of that code has been put in the public domain. Making something available for download is not the same as putting the associated copyrights or intellectual property in the public domain.

As to which you should "look for," it depends. If you're still talking about GXemul ( viewtopic.php?f=18&t=16729104 ) then you don't need anything from here.

If you're talking about MIPS-based DECstation hardware, none of them - those are all for Alpha-based systems.

If your "DECstation" is actually a DEC 3000 model XYZ , AlphaPC, Personal WorkStation, or an AlphaStation 2xx/400/500/600, then you might be able to find a firmware update for that machine under ftp://ftp.hp.com/pub/alphaserver/firmwa ... _platforms .

smj wrote: [ In the early 80s you're looking at top-of-the-line hard drives of a few hundred megabytes, using 14" platters and interconnects like SMD, and weighing hundreds of pounds. An example common by the mid-80s would be the Fujitsu Eagle , a ~470MB SMD drive using 10.5" platters that would retail for $10,000. Meanwhile in 1985 Seagate announced the ST4051, a 40MB 5.25" HDD which would initially cost you almost $1,000 in OEM quantities, let alone retail, and provide such slow access times and throughput that the same workstation running diskless could often out-perform it despite having all disk I/O go over 10Mb Ethernet to a fast server with an Eagle.

I had a couple of K&T's from the mid-seventies that had Shugart disks. 19" rackmounts with maybe a 14" single disk, about one meg ? It was a big shiny copper-plated thing. You could comb your hair in it ... Belt drive, to change from 60hz to 50hz you changed the pulleys and belt They had a transparent plastic cover so you could watch the heads moving. The weight was maybe 50lbs. They were more reliable and cheaper ($5,000 maybe ?) than core memory but core was way faster. 32k of core was several thousand from aftermarket guys. There was a nutcase in Texas - Federated Computing, maybe ? - who had a warehouse packed worse than mopar's house with DEC stuff. He had everything and a lot cheaper than DEC prices. God knows what core cost from DEC. Only He would have the testicular fortitude to ask.

Funny thing about those old controls : the hardware was awful but the software was great. Now the hardware is great but the software sucks the big ten inch. Somewhere, something went wrong with American education

Larger 5.25" HDD were available, and their prices did drop rapidly in the latter half of the 80s, but it took a while for the performance equation to change.

By the mid-eighties SCSI must have existed ? K&T had a kit to replace the older Shugarts with a scsi drive, but the kit cost more than my car. Then K&T went broke. Hmm, any connection there, do you think ?

Along those lines, I remember Bendix wanting $ 7,000.00 USD for threading macros for a 5M control. That's not a misprint : seven thousand 1976 dollars. For a crummy little macro. David Love bought a used Testa Rossa for that, right about then. You could buy two Dodge Challenger R/T's for that. Or make the downpayment on a Marin county house. It was utterly ridiculous.

Bendix is also gone now

hamei wrote:

Larger 5.25" HDD were available, and their prices did drop rapidly in the latter half of the 80s, but it took a while for the performance equation to change.

By the mid-eighties SCSI must have existed ?

Yes, you had 8" and 5.25" drives with SASI (pre-SCSI) interfaces from the start of the decade, and SCSI became more common as the decade progressed. SCSI was commonly used for desktop workstations, and often the deskside models, but each manufacturer made different choices over the years. You might find native ESDI controllers in deskside machines like the IRIS 3000 line, MFM drives in early DEC MicroVAXen (pedestal/deskside), or with Sun you'd find SCSI controllers in the host with adapter boards like the Emulex MD21 or Adaptec ACB4000 that spoke SCSI to the host and controlled MFM or ESDI drives until mid-decade.

The 5.25" drive mechanisms only had 1/8 - 1/4rd the throughput of the big drives to start with. Translation like Sun's use of the SCSI - ESDI/MFM boards rarely helps performance, but you were often stuck waiting for the drive anyway. On the other hand, if you were only buying one or two systems at a time, you needed local disk - tons of "standalone" Sun-3's were sold with 70, 140, and 300MB drives in a shoebox, or inside the deskside models. Naturally your trade-offs and budget were different if you were buying a dozen workstations and a server...

The later you got in the decade you saw faster and bigger native SCSI drives from Micropolis, CDC/Imprimis, etc. Quantum started with 8" drives and continued down through 5.25" to 3.5", and probably supported every interface along the way... Naturally the equation changed if you were shopping in 1984 versus 1989. And once you could get decent 3.5" drives, you saw the typical 90s-style pizza box emerge that could hold one or two internal drives, and things changed again. The VAXstation 3100 line shows up around 1988, the DECstation 3100 and SPARCstation 1 show up in 1989 IIRC.

And then there was IPI-2 which was positioned at the high end. My 1990-vintage 4D/380VGX PowerSeries has two SCSI channels (IO3 board), used for QIC150, CDROM etc, and a Xylogics SV7890 VME IPI-2 controller. In the bottom of the rack sits an 8" Seagate ST81154K (a.k.a. CDC 'Sabre'), which basically eats up half a 19" shelf, consumes 160W (it has it's own 220VAC PSU) and weighs roughly 30kg. It has it's own key panel and LCD screen at the front which you can use to manually key in bad sectors, I believe. Oh, the raw capacity is ~ 1.1GB. It's a bit of a monster. And it still works

IPI-2 has more bandwidth than the synchronous SCSI-1 implementation of the PowerSeries. The ST81154K internally works like a RAID0 (stripe) so it's streaming speeds are faster than you might expect. The PowerSeries PROMs can boot from the Xylogics controller, and 'fx' can format it so you can use it as an IRIX system disk.

IPI-2 had some other unusual features, I think you could attach a disk to two systems for fail-over, similar to Fibre Channel. But that pretty much sums up what I know about IPI-2. Information about it is pretty scarce.

hamei wrote: By the mid-eighties SCSI must have existed ?

Yep. SCSI was one of the selling points of the Mac Plus back in 1986. You could add 10-20 MB for around $1,000, IIRC...and it was worth it, too!

Hey, I remember the topology of the setup now because I read back my saved email in 1995.
There were several DECStations sitting in the lab, and one OSF/1 data server at a remote site.
But One doubt is where did the ultrix reside? at the server I think because the DECStations were diskless.
So maybe the OSF/1 server cluster delivered the Ultrix OS thru DECnet to the DECStations when being logged in.
Thanks
Jack

the ultrix kernel would have been stored on the Alpha server. DECstations were capable of two different methods for netbooting: a) using Internet protocols bootp and tftp, or b) using a protocol from the DECNet suite called MOP. They do exactly the same thing, but if your network infrastructure was heavily DEC based you might have used MOP, versus bootp/tftp in a TCP/IP network.