A WORKSTATION RANT
By
Peter N. Glaskowsky [png@ideaphile.com]
Previous works by Peter Glaskowsky available on Chaos Manor include mail and two special reports: A Report on the Meltdown Conference, and a special report on the Casio Cassiopeia. Robert Bruce Thompson's report on NT Backups inspired another Glaskowsky piece, which is either a manifesto, a rant, or a special report, whichever you choose. Whatever you call it, it's interesting. I've added some correspondence to the end, and one day I'll try to edit this for formatting, but it's readable as is and that's in the important thing. JEP
Robert Bruce Thompson’s report on NT backups reminds me of a rant that is turning into one of my signature pieces. I wrote up these thoughts first in the form of a Technology Roadmap report for MicroDesign Resources, but that was a while back. I’d like to offer an updated and condensed version.
Here’s what set me off this time. Thompson points out that SCSI-based tape backup drives have a hidden cost for the SCSI adapter itself. He says "most workstations" don’t have a SCSI adapter. I don’t think this is quite true, even for PC workstations, but the fact that there are _any_ PC workstations without SCSI adapters shows that there’s a real problem with how workstations are defined today.
The best example I can offer of how the term "workstation" has been systematically rendered meaningless is Hewlett Packard’s new Kayak XA workstation. These are very inexpensive systems. They sell for as little as about $1904 (see http://www.hp.com/kayak/price/xa-price.html ). At this price, you get a 350-MHz Pentium II processor, 64M of RAM, a 4.3G Ultra-ATA hard disk, a 32X IDE CD-ROM, a Matrox Millenium G200 graphics card with 8M of RAM, and NT Workstation 4.0.
When you buy this base-model HP Kayak XA, you don’t get SCSI. You don’t even get a LAN adapter. Even if you buy the best Kayak XA they make, you’re not really getting a workstation, you’re getting a generic PC. Why does HP call it a workstation? Because that lets them sell these system for hundreds of dollars more than an equivalent HP Brio business desktop. Naive customers think they’re getting something better than just another PC, but they’re not.
This isn’t really fraudulent, since there’s no such thing as a minimum configuration for a "workstation". It’s still a tragic waste of a perfectly good word that used to mean something.
A "workstation" used to be a high-end personal computer meant for use by a professional engineer, designer or programmer, typically in the mechanical, electronic, graphics, or database-management industries. Such systems offered high performance at a high price; while they couldn’t match the price/performance of mainstream PCs, they could handle tasks that mainstream PCs simply couldn’t. Historically, they had far better floating-point performance, greater main-memory bandwidth, and much more scalability. Some traditional workstations still surpass PCs in all of these categories (and price).
Even the first workstations included integrated connectivity for workgroups. They came in an unusually wide range of compatible systems so that companies could find a good match between the demands of its users and the features of a specific system. They were equipped with a multitasking operating system to get the most out of the hardware—background tasks could use up any spare cycles while the operator performed foreground tasks. They had peripherals consistent with these goals—premium graphics controllers suitable for the target application, SCSI adapters to provide high-performance mass storage with multitasking support, the fastest available networking cards, etc. These peripherals resided on proprietary expansion buses with higher effective bandwidth than anything you’d find on a contemporary PC.
Perhaps most importantly, workstations used to be designed to a higher standard of quality. They had higher hardware costs because they used more expensive IC packaging and expansion-bus connectors to improve electrical signal quality through reduced lead inductance and a higher ratio of power &; ground pins to signal pins. They had bigger heat sinks, more reliable power supplies, heavier cases, and other features that improved hardware reliability and made these systems sensible choices for mission-critical tasks where failure could be tremendously expensive.
Workstation operating systems and applications have traditionally received special attention, making them similarly more expensive and more reliable, and even today they have features that go beyond anything Microsoft is shipping, such as 64-bit addressing, scalability beyond 8 processors, and advanced user-interface technology.
The HP Kayak XA (and similar models from Dell and Packard-Bell/NEC) utterly fails to measure up. There’s just no sense of premium quality in these machines. Yes, they have some of these characteristics: NT Workstation is better than Windows 98 for mission-critical applications, and it supports networking. The case uses heavier steel than a regular PC. This just isn’t enough to justify the workstation name.
Buy any generic PC, slap in a 3D-graphics card, a network card, IDE storage devices, and NT Workstation, and you’ve still got a PC. The hardware is no faster and no more reliable than a PC. The OS is only slightly more reliable than a generic desktop OS, and while it may be compatible with multithreaded software, the overall system provides no specific support for multithreading.
What makes a machine like this a "workstation"? Nothing. So why call it one? It isn’t only the PC workstation vendors who are misleading their customers this way. Sun calls its Ultra 5 a "true workstation", but this is a true lie—the Ultra 5 is basically a PC clone with an UltraSPARC processor grafted on. It uses PC-class graphics and an IDE hard disk. It has the same kind of ASICs and circuit boards you’ll find in a generic PC, with no better reliability. Sun certainly knows how to make "true workstations", but the Ultra 5 doesn’t qualify.
Admittedly, it’s getting harder for the traditional workstation vendors to compete with the PC workstation vendors. Intel’s Pentium II and Pentium II Xeon processors are faster than the best offerings in some RISC families. SCSI and Fast Ethernet work just about as well on the PC platform as they do anywhere else. These are the hooks upon which the "PC workstation" label is hung. But is that all there is to it? Hardly. There’s still plenty of room in this industry for premium systems that make professional users more productive than they’d be with a generic PC. It’s even okay if they’re not PC compatible, as long as the lack of compatibility permits special features that you can’t get in a PC.
We need a new word, I think, to make it easier to identify these premium systems. I just can’t think of a way to keep unworthy OEMs from stealing any new term we might come up with. Ideas are welcome.
. png
Send your new ideas and I'll include them either here or in mail. JEP
Peter,
Sorry to add to the confusion. In context, I was using "workstation" in the sense of a "Windows NT Workstation client computer" rather than in the traditional Unix workstation sense. In that sense of the word, "most workstations" indeed do not include SCSI - probably 99.9%+. Ordinarily, I would have used the more accurate term "client PC", but WinNTMag supposedly wanted an article on "workstation backup", whence my choice of terms.
I enjoyed your rant/manifesto/special report, and agree with everything you said.
Regards.
Bob
Robert Bruce Thompson
Second Thoughts by Robert Bruce Thompson
<<Buy any generic PC, slap in a 3D-graphics card, a network card, IDE storage devices, and NT Workstation, and you’ve still got a PC. The hardware is no faster and no more reliable than a PC. The OS is only slightly more reliable than a generic desktop OS, and while it may be compatible with multithreaded software, the overall system provides no specific support for multithreading.>> Then again, what would you call a system built on a PC Power &; Cooling tower case with a 600 watt power supply that contains an Intel dual-processor Pentium II system board with two 400 MHz PII processors, 768 MB of 100 MHz RAM, Fast/Wide SCSI and 100BaseT on the system board, 54 GB of disk, a 24 GB DDS3 tape drive, AGP-2 video card with 16 MB, a 21" monitor, and dual-boots Windows NT 4.0 and Linux or BSDI?
It’s a PC, certainly, but it compares favorably in every respect (including construction quality, robustness, and reliability) with many systems that I’m sure you would grant the official "workstation" label, including boxes that would have been top-of-the-line workstations not all that long ago. Or does today’s workstation lose its "workstation" status as it ages? I’m not sure that any rigid line exists nowadays to differentiate workstations from PCs. As Oscar Wilde said, "One man’s Mede is another man’s Persian."
Regards.
Bob
Robert Bruce Thompson
thompson@ttgnet.com http://www.ttgnet.com
To: <thompson@ttgnet.com>
From: "Peter N. Glaskowsky" <png@ideaphile.com>
Subject: Re: Second thoughts on workstations
Date: Tue, 8 Sep 1998 19:07:27 -0700
>> Buy any generic PC, slap in a 3D-graphics card, a network card, IDE
>> storage devices, and NT Workstation, and you’ve still got a PC. The hardware
>> is no faster and no more reliable than a PC. The OS is only slightly more
>> reliable than a generic desktop OS, and while it may be compatible with
>> multithreaded software, the overall system provides no specific support for
>> multithreading.
>
> Then again, what would you call a system built on a PC Power &; Cooling tower
> case with a 600 watt power supply that contains an Intel dual-processor
> Pentium II system board with two 400 MHz PII processors, 768 MB of 100 MHz
> RAM, Fast/Wide SCSI and 100BaseT on the system board, 54 GB of disk, a 24 GB
> DDS3 tape drive, AGP-2 video card with 16 MB, a 21" monitor, and dual-boots
> Windows NT 4.0 and Linux or BSDI?
Well, I’m willing to call that a "PC workstation." It still has a PC-class processor (so there’s no premium value there), but it has two of them. It probably uses Intel’s best PC core logic rather than Intel’s workstation chip set (the 440GX), so it isn’t as fast or as reliable as what you’ll find in an SGI workstation but is still pretty good. The SCSI &; Ethernet controllers are stuck on the mainstream 133-Mbyte/s PCI bus, which really doesn’t have room for anything else, but that only impacts future scalability to Ultra2 SCSI and Gigabit Ethernet. Peripherals are a wash as far as I’m concerned; they don’t affect the nature of the system itself. Those are good operating systems but they just aren’t as sophisticated or mature as Solaris or Irix. On the other hand, NT has a huge edge in application compatibility. The better case and power supply help, but without a high-reliability chip set and motherboard there’s only so much those things can give you.
I believe the first true x86 workstations are yet to come—systems from traditional workstation vendors, designed from the ground up around Intel’s new Pentium II Xeon processor (and Merced, later) and proprietary chip sets. SGI itself will be represented in this category, and insists its systems will be the best in the world when they appear Real Soon Now. HP and IBM will offer stiff competition. Compaq will use its Digital connection to try to gain a share too; its own PC workstations have been pretty decent machines, but they’ve left room for improvement.
> Or does today’s workstation lose its "workstation" status as it ages?
> I’m not sure that any rigid line exists nowadays to differentiate
> workstations from PCs. As Oscar Wilde said, "One man’s Mede is another
> man’s Persian."
Well, that’s the essence of the question I asked in my rant. Is there any such line? Sometimes I think of it as "anything too good to be a PC," and some of these things are timeless. Motherboards aren’t gaining layers over time; sheet metal doesn’t get thicker. Some of them expire quickly, of course, like CPU speed grades—so maybe these factors are irrelevant.
Thanks for the comments!
. png
----------------------------
From: "Robert Bruce Thompson" <thompson@ttgnet.com>
To: "’Peter N. Glaskowsky’" <png@ideaphile.com>
Subject: RE: Second thoughts on workstations
Date: Tue, 8 Sep 1998 22:54:36 -0400
> Well, I’m willing to call that a "PC workstation." It still has a PC-class
> processor (so there’s no premium value there), but it has two of them. It
> probably uses Intel’s best PC core logic rather than Intel’s workstation
> chip set (the 440GX), so it isn’t as fast or as reliable as what you’ll
> find in an SGI workstation but is still pretty good. The SCSI &; Ethernet
> controllers are stuck on the mainstream 133-Mbyte/s PCI bus, which really
> doesn’t have room for anything else, but that only impacts future
> scalability to Ultra2 SCSI and Gigabit Ethernet. Peripherals are a wash as
> far as I’m concerned; they don’t affect the nature of the system itself.
> Those are good operating systems but they just aren’t as sophisticated or
> mature as Solaris or Irix. On the other hand, NT has a huge edge in
> application compatibility. The better case and power supply help, but
> without a high-reliability chip set and motherboard there’s only so much
> those things can give you.
Yes, and I kind of cheated on the OS MP support, too... I wasn’t really trying to argue that you can build a true workstation around the PC architecture, just that you can come mighty close for all intents and purposes. And you can do so for a lot less money. If money’s no object, then of course you go with the true workstation. However, the low cost of "super-PCs" has allowed a lot of people to have pseudo-workstations on their desks who wouldn’t otherwise have that much. Let’s not discount the downward price pressure that high-end PCs have put on true workstations, either. Without the high-end PC, I doubt the true workstations would be anywhere near as affordable as they are now.
> I believe the first true x86 workstations are yet to come—systems from
> traditional workstation vendors, designed from the ground up around
Intel’s
> new Pentium II Xeon processor (and Merced, later) and proprietary chip
> sets. SGI itself will be represented in this category, and insists its
> systems will be the best in the world when they appear Real Soon Now. HP
> and IBM will offer stiff competition. Compaq will use its Digital
> connection to try to gain a share too; its own PC workstations have been
> pretty decent machines, but they’ve left room for improvement.
I think you’re probably right, although I’m not really that impressed with the Slot 2 processors. Still, they’ll allow people to build boxes that provide about 95% of a workstation’s functionality for a small fraction of the price. I think perhaps you are giving too much weight to reliability, also, or rather to the differential in reliability. PCs are an order of magnitude more reliable than they were a decade ago. You can get PCs, for example, with redundant power supplies, RAID, ECC memory, etc. etc. For something like air traffic control, I’d rather have a workstation. But for routine workstation tasks, I really don’t think there’s all that much difference nowadays on the hardware side. Software, certainly.
> Well, that’s the essence of the question I asked in my rant. Is there any
> such line? Sometimes I think of it as "anything too good to be a PC," and
> some of these things are timeless. Motherboards aren’t gaining layers over
> time; sheet metal doesn’t get thicker. Some of them expire quickly, of
> course, like CPU speed grades—so maybe these factors are irrelevant.
I think that each class of system is benefitting from the other. A lot of workstation technologies are becoming mainstream PC technologies, and workstations have benefitted from the economies of scale that PCs yield. If it weren’t for PCs, workstations would probably still be using 10Base5 coax, discrete memory chips, tiny little hard disks that cost $10/meg, and so forth. I think the two classes will ultimately merge, with workstations contributing the high-end technology to raise the bar from that end and PCs the mass production needed to lower the bar on costs. I’m sure there’ll always be a market for high-end workstations, just as Rolls Royce sells a certain number of cars every year. But I do see the gradual merging of the two continuing.
> Thanks for the comments!
Thanks for your thoughtful posts. I enjoy each of them.
Regards.
Bob
Robert Bruce Thompson