Here's a blinding insight for you: Moore's law applies only to hardware, not to the total cost of systems.
What sparked this recognition of the glaringly obvious was a review of Apple's new Xserve (a dual processor 1U blade server) and some thinking about what that thing is likely to mean for the Linux marketplace.
Although the actual numbers are bit fuzzy, it seems that Sun leads the Unix market in terms of the number of users served, IBM leads in Unix related revenues, and Apple, not Dell or HP, sells the most Unix boxes. Those machines run the MacOS X layer on top of Darwin, an opensource BSD variant with a MACH kernel.
Technically the most interesting thing about Apple's Unix is the use of the PDF imaging model; itself derived from the Postscript imaging model from NeXtStep, within the MacOS X layer instead of the more traditional X based approach. With Darwin you can add X11 at any time, and even run both concurrently, but the imaging system defaults to the Postscript derived PDF display model.
|
Sun pioneered
this many years ago with NeWS, but ran into rights
related issues that drove up the cost, and limited adoption, at just about the same time
that MIT was giving away X. Brain damaged or not, free generally beats elegant and I'm typing
this under X, not PDS.
On the other hand, Solaris still provides native postscript display support, including the use of trutype fonts, via the Display Postscript (DPS) extension to the Solaris X server. There's a group working on supporting this in the Xfree86 community too. Linux utilities like dpsexec work well but I didn't get my full screen display to work; probably because I simply didn't know how. Adobe also continues to support display postscript, providing the toolset and support through various partners. One effect of Apple's market success will, I expect and hope, be to revitalize interest in PDS technologies and such associated goodies as the GNUstep initiative to provide the benefits of NeXt's development environment under Linux. |
Like Linux, the underlying Unix for MacOS X is an open source production and, again like Linux, it has all the traditional Unix virtues including high reliability, network compatibility, efficient resource use, and access to a wide variety of lower cost, cutting edge, tools and applications.
The desktop MacOS X shell is a revenue generator for Apple, but Darwin is freeware and so people buying Apple's Xserve product are not being charged separately for the OS. The result is a Mac server that costs about the same as a Linux server and, therefore, considerably less than a Windows server. What's particularly interesting about this is that it contradicts some common wisdom: Intel boxes are cheaper because so many are made, right?
Sun's lowest end gear, the V100, costs more than an x86 machine with similar performance but, at $995 list in lots of one, it's not intended for much more than certain high volume embedded uses where SPARC compatibility (particularly the VIS/SIMD stuff) is more important than cost or performance. The V480 and V880, however, outperform their x86 counterparts for less money and I had been assuming that this reflected Sun's willingness to take a hit in these markets in exchange for better sales in the more profitable mid range markets. Apple's Xserve, however, made me rethink those assumptions because it too is cheaper than comparable x86 boxes running a Windows OS and pretty clearly isn't intended as a loss leader.
What I now think is going on is that Moore's law has gotten itself transmuted to Unix software while the proprietary systems vendors have gone the other way. Consider that:
Ignoring both the inflationary decrease in the value of the dollar and increases in computing power and utility, these changes can be summarized as:
| Approximate Cost | ||||
|---|---|---|---|---|
| Hardware | Microsoft Operating System | System Total | ||
| Desktop | 1981 | $2,959 | $39.95 | $3,000 |
| 2002 | $450 | $199 | $649 | |
| Percent Change | -85% | 500% | -78% | |
| Server | 1985 | $12,000 | $495 | $12,500 |
| 2002 | $5,877 | $3,295 | $9,172 | |
| Percent Change | -49% | 655% | -36% | |
When you don't pay for the operating system directly you get the full benefit of gains in hardware design and manufacturing --that's why you can get a Lindows machine at Walmart for less than the $450 shown here and pay only $2,953 for the base Dell server if you choose to run Linux or BSD on it.
Apple's Xserve follows this paradigm: stripped down it costs marginally more than roughly comparable Dell x86 hardware with Linux and substantially less than roughly comparable Dell x86 hardware with a Microsoft OS.
| Apple's Xserve (Darwin) | Dell 1650 with Red Hat | Dell 1650 with Windows 2000 (25 clients) | |
|---|---|---|---|
| 1 CPU, 256MB, 60GB | $2,999 | $2,953 | $6,079 |
| 2 CPUs, 512MB, 60GB | $3,999 | $3,896 | $7,022 |
| 2 CPUs, 2GB, 480GB | $7,799 | $10,334 | $13,495 |
| Apple's big cost advantage on their "large" model comes from the use of 120GB ATA drives instead of SCSI drives. A Dell PowerVault Ultra160 setup to match 480GB would add about $11,000 to the cost but mis-represent the performance difference, so for costing I used 5 x 73GB disks at $699 each and pretended to be able to put them internally on the 1650. | |||
Performance is, of course, a hot issue here because the entire systems architectures, from CPU type to operating system file management, are very different between these two classes of machine. Recent benchmarks by xinet show the Apple server well ahead of the Dell running Windows 2000 but comparable results for that machine running Linux are not yet available and the performance comparison, particularly on less obviously Mac oriented tasks, will almost certainly be a matter of dispute for some time.
From a cost perspective we need to ask the question: what happened to returns to scale here? Companies like Dell and HP ship something like eight or nine x86 machines with some kind of Windows OS for each Unix machine shipped by anybody --including their own Linux shipments as well as both Sun and Apple. Yet the lower volume product costs less and arguably does more. Adam Smith would not find this intuitively obvious, so what's going on?
The raw hardware costs for Sun, Apple, and Dell all seem to be very nearly the same for approximately comparable performance levels despite enormous differences in production volumes. The major pricing differences in the machines as shipped are not coming from production cost differences, they're coming from add-on license costs. Unix has been getting cheaper right along with the hardware while Windows has been getting more expensive. Run 100 client licenses on that Dell server and the software will cost substantially more than the hardware, but put Linux on that machine and the cost doesn't change whether you have one user or 100.
Consciously or not, Linux and BSD users have known this for years, but we've been voices in a wilderness. Telling the boss that Unix is cheaper and better without both giving him a handy headline to hang his understanding on and having someone else reinforce the message independently, just hasn't been that effective. It's frustrating but true, a lot of people who've given my defenestration guide to their bosses have told me that these people often won't read or understand it, in large part because they feel no social pressure to do so.
This is where Apple's new products will help. That Xserve demonstrates both the cost advantage of the open source model and the network advantages of Unix to a far wider audience than we've ever had before. Given that the OpenOffice.org people now have the developer release of their suite running under MacOS X, we can expect, furthermore, to see that message explode out of the server rooms and onto the desktops of people who, as a group, have had to endure a lot of social pressure against their choice of the Mac even while, or perhaps because, Microsoft struggled to catch up to the original and now obsolete, MacOS.
Apple should ship almost four million Unix desktops this year, and each one of them represents a new opportunity for open source ideas to take root and for products like OpenOffice.org to find users. Equally importantly, each time a Mac moves into an office environment it gets harder to maintain the fiction that homogeneous (meaning all Windows) systems are cheaper or easier to run. Correspondingly, it should get easier for people to use Linux or BSD to extend the life of existing servers --and thus better serve both their Windows and Mac user communities while reducing corporate data processing costs.
Linux, of course, isn't just a server --that's a distinction reflecting non Unix technology-- it's also a powerful desktop solution. The new Macs illustrate this joint capability (you can turn the Xserve into a workstation by replacing one network card with a graphics board) and you would therefore expect to see people now using a few Macs in an office to start thinking about the advantages of skipping Microsoft Licensing 6.0 by transitioning users to Linux on existing x86 machines. After all, not everyone needs the full MacOS X shell and companies can get most of the user and management benefits Unix offers by simply re-using existing Windows hardware.
So is this happening now? Although the definitive results won't be in a for a year or two, an editor at macslash.org put up a request from me asking their readers for comment. This produced quite a lot of thoughtful comment and email, much of it suggesting that some of these people are seeing changes in the behaviors of people who would previously have acted as if non Windows choices simply don't exist.
All of these are, I think, early warning signs that the social controls whose operation has kept so many users and executives in the dark about the real costs and consequences of making an all Windows decision are weakening dramatically. When that dam breaks (assuming it does) the repercussions are going to be interesting. Who, after all, would put in rackmounts of Windows servers where a mid range Unix machine would do the same job for much less, if the social pressure to do it disappeared?
Overall, is this good news for the Linux community? I think so. From a purely business driven perspective, four million new Unix users a year is a market that's ours to lose. From the more personal and interesting perspective of Unix evangelism, these are the guys to have on side because they demonstrate that computers can be used in quiet, effective, ways that contribute to the organization and don't require inordinate amounts of support.
It is past time to welcome the Mac community into the general Unix and open source movements; after all, we're all working to expand and apply the same set of ideas, with the same set of tools, and for the same personal and business reasons. So take a Mac user to lunch and talk about like how easy it is to integrate the new Macs on your Linux server network or how you've defaulted your departmental printers to Postscript specifically to recognize the Mac graphics advantage. Making new friends for yourself and for Linux here is easy: just make it clear that you're on side with them, and they'll get on side with Linux, and with you.