Sorry for the interlopage, this is in fact my first post here and one of few to SI at all. Not to disagree with anyone, I thought that I'd add just a little insight from the customer perspective; I've been buying Suns for my employer for over 12 years, starting with the 2/130 in 1985.
The $2,995 Ultra 5 is not meant to be chosen and purchased by an individual or end-user organization: Rather, it is a bid box. It is there for the convenience of systems integrators who bid on government and large corporate RFPs. It serves the purpose of lowering the overall cost-per-seat in a proposal. In this use, it makes complete sense that the unit would not come with CD, floppy or monitor; the integrator is going to want to buy bargain units and stuff the box anyway. It also somewhat conviently results in change orders after contract award. Previous bid boxes from Sun include the 4/110, ELC, IPC, Classic and, most recently, the SPARCstation 4.
A secondary purpose is of course to allow for press releases that use words like "Sun workstations for under $3000". Thus, it is very much like an "ad car" in the automobile business: No radio or air, power nothing, single-speed wipers, optional back seat, but alas, the ad can read "Only $7,999!". Show up to buy one and the sales guy yells to the back "Hey Joe! Did we sell the ad car yet?"
To some extent, this latter purpose can act against them. You can see that in the derision here for the bottom end Ultra 5. Look instead, however, at something that someone would actually choose to buy, like the Ultra 10 with 128MB of memory and the Elite-3D graphics. This beast has a SPECfp95 of 12.9 and can crank 3 million 3D triangles per second, all for $10.4K list. To the extent that you can begin to touch this with a PC, you're looking at something like the Compaq M6300, which sells for about $7.7K street. But the processor/memory bandwidth, the floating point performance, and the graphics performance of the Compaq all suffer in comparison to the Ultra 10, which is only one step above Sun's entry level Ultra 5. Look at the Ultra 60, with two 300MHz processors, 512MB of memory, an MP Specfp95 of over 23, and six million 3D triangles per second, all for a mere $27K. Yeah, it's a lot of money, but please find me an Intel-based platform that can do these things at all. And if that turns out not to do it for you, you can always upgrade to a six-processor Ultra 3000 workstation configuration with 6GB of memory and 90GB of internal disk. With the Compaq, there's simply no place to go. Seriously, though, we are talking movie studio, Wall Street or NSA-style applications here.
I have purchased bargain Sun bid boxes several times over the years, but rarely as actual desktop workstations. I've used them most often either to give to our systems guys for prototyping our OS configuration, preparing disk drives for production machines, reproducing odd problems in a test setting for diagnosis, etc., or as low-end servers. We use classics, for example, for an internal anonymous ftp server, to run the software license managers, and as a dedicated time server hooked up to a TrueTime GPS reciever. For these uses, typically you don't need a floppy, CD-ROM or monitor; in fact these parts are typically a liability for such applications. And having them near the bottom of the performance rankings reduces the propensity of the user community to attempt to run analytical jobs on them :-)
In fact, my biggest gripe with the Darwin announcement is that they didn't offer an explicit server configuration (the primary issue being the OS license). I'd really like to take about 30 of those Ultra 5s, put them on shelves in a few EIA machine racks, and let our users run their long-running econometric models there. Since many of our jobs are extremely processor/memory intensive, the context-switching rate and locality-of-reference assumptions that go into designing the big multiprocessor systems go straight out the window; we do much better with an array of uniprocessor units each with dedicated DRAM. If you read through the UltraSPARC III announcement, you'll see that it included a provision for per-processor DRAM in addition to per-processor E-Cache. I don't honestly believe that I had an effect here, but for the first time in quite a while I felt like Sun had been listening to my screaming about this issue.
We haven't extensively used Suns as desktop workstations since 1990; they are too expensive for that [we really don't need some of their advantages on the desktop] and it is just too difficult to maintain that many Unix OS images (we have about 350 users). We have used NCD X terminals for years. Although they are brain-dead easy to manage (no disk drive, no pain), they too just fell behind in features and cost. The 8-bit color issue is probably the most painful in an age of Netscapse and WordPerfect, but also they don't support virtual memory, so they keep eating ever more semiconductor memory to support the newer releases of software (Java-based NCs have this same problem, BTW, only in spades: They must store not only display data but also executable code in semiconductor memory). Thus we too are migrating to PCs running windows NT. We are deploying PCs with 64-128MB of memory, 21" apeture-grille monitors, and graphics at 1600x1200, 64K Colors @75Hz, and use eXceed for the X server. To approach this performance level in an X terminal or a Sun workstation simply costs a fortune by comparison. [We use the STB Velocity 3D w/4MB right now, based on the S3 Virge/VX. This year we will probably switch to the Permedia-2, or a 3DLabs-based ELSA for high-end requirements].
But the wonderful thing about the Suns is their unmatched stability. I was having a problem the other day getting my email from the POP server running on my own server, and I started to suspect the Sun.
This was a mistake -- it turned out to be the NT client software. But that Sun (a SPARCstation 5) had been up, I noticed, for 47 days, and even then there was really no reason to restart it. I cannot begin to count the number of times I had rebooted my NT machine since then. We have users that depend on the ability of the Suns to run jobs that take up to *two months* to complete. I simply cannot imagine migrating this kind of task to an NT workstation.
True, our users will on the NT machines have decent spreadsheets, word processors and GUI-based relational databases for the first time, and I expect that some analytical work will migrate to these machines in due time. But I simply can't imagine the Suns going away any time soon. I just need to find a way to get these Darwin machines into a rack; there are in fact some real bargains there.
--Bob Drzyzgula |
