|HOME||PUBLIC LIBRARY||ANOTHER PERSPECTIVE||INFOPERSPECTIVES||CONTACT|
It's a shame Linus Torvalds, the father of Linux, didn't come from a city called Milan; he hails from Helsinki. Milan, Italy, is where Leonardo da Vinci spent some of his most productive years. Milan, Ohio, is where Thomas Edison was born. Da Vinci and particularly Edison were good at exploiting their inventions for commercial ends, while Torvalds has left the exploitation to the distribution and support outfits. The result has been a regressive economic structure, rewarding populists with cheap, standardized X86 and now X64 systems based on Intel's concepts, while punishing those who are wedded to proprietary hardware.
There is nothing sinister about the way the Linux market has evolved. It's the natural consequence of market forces. Technologies that once offered cost advantages simply no longer do so, not only for Linux users but also for everyone in their market segments.
This is what happened in both Milans. Milan, Italy, once had a wonderful canal system and became a leading port in Europe even though it is inland. Most of the system is now covered over or filled in. The pieces of the canal system that are still left have yielded an urban marina that's one of the liveliest districts in the beautiful old city. The old hidden canals, some of them built using lock technology falsely ascribed to the inventive mind of da Vinci, but in fact first constructed before Leonardo was born, were largely forgotten by the Milanese until last month, when the Science Museum in Milan took delivery of a small submarine.
The submarine is going on exhibit to remind people of the many inventions thought up by da Vinci, which include flying machines and a lot of military apparatus. It was much harder to come up with such ideas in the fifteenth century, because the light bulb for the thought balloon wouldn't be invented for centuries, when Edison got around to creating a gadget that might have gotten a lot of wear and tear if it had been around in the time of the other man of Milan.
It cost a million euros to schlep the boat into Milan because the city had to remove and restore overhead lights, cables, and power lines, widen some roads and reinforce the pavement where modern thoroughfares have been built atop the old canals. Imagine that: so much work to keep a submarine from getting wet.
Milan, Ohio, was a canal center, too, in the nineteenth century. It was, in the mid-1800s, such an important railhead that it claimed to be the second busiest grain port in the world. Odessa, in Russia, was number one. Today fewer than 1,500 people live there. They can walk the former canal, if they want; it was closed down in 1868 after too much of a good thing, in this case a flood.
Transportation technology moves on, leaving wonderful relics and nice memories, and there's not much that can be done to hold back the change. This turns out to apply to the transportation of data as well as wheat.
Love it or loathe it, computing is moving ahead, and the present phase seems to involve a shift to processing on the X86 galaxy of chips that are either made by Intel or made from Intel's ideas. This year, for example, Apple Computer, an Intel (or at least IBM PC) basher in its youth, passed the X86 milestone The maker of the Macintosh, which has by far the highest unit volume sales of any proprietary computer vendor, has thrown in the towel; it's going to move from Power chips to X86 chips. But Apple not going to give up its proprietary systems software, any more than Sun, now offering X86 servers alongside its line of Sparc machines, is about to abandon its Solaris software, or HP ditch its HP-UX Unix variant.
The most complex situation is that of IBM, which makes a lot of proprietary iron but backs Linux everywhere, too. IBM's X86 servers are natural platforms for the most popular versions of Linux. Its two Power-based product lines, the i5 and p5, each have their own operating systems but also can support Linux. IBM has made a special effort to promote Linux on its mainframes, customizing its mainframe engines with microcode to allow users of the multiprocessor computers to, in effect, run a server cluster inside a single frame.
Despite all these vendors' efforts, though, the fact remains that no proprietary hardware, produced in modest volumes, can provide the value of X86 servers, made by the millions. Other architectures have to compete on grounds other than raw hardware cost. Some do this by providing more power per engine, some by offering superior multiprocessing technology, some by promising faster communications with storage devices and other peripherals, some by giving better protection from component failures. But for many users, the most prominent economic reason for sticking with a proprietary system is an investment in programming that can't be preserved if the proprietary system is dropped in favor of an X86 alternative.
What some users with proprietary systems have tried is to move to mixed environments. Those with IBM machines may have the easiest time, because the first step to mixed environments can be done with software alone. IBM's Power-based and mainframe processors can run Linux alongside other operating systems, giving users a chance to try out Linux without buying a rack of X86 servers. Still, there are complications, and it may well be the case that users who want a Linux box should just go out and get one.
For most users of smaller IBM mainframes, machines with power under 100 MIPS, migration has become a way of life. These customers often don't move all their work at once. Instead, they typically start by running email on an X86 box and then maybe a web site. After that they may check out moving a cluster of applications, such as human resources systems, off the old iron. Ornate, customized accounting systems are harder to move, and the same goes with other complex suites, such as ERP. Once the users start to shift their work, they generally don't need bigger mainframes, they need even smaller ones, but these days IBM offers only big and costly systems. The big IBM machines can be throttled back to slower speeds (and lower price tags) only by disabling cache memory or otherwise hobbling the native potential of the mainframe engines.
For these users, putting Linux on a mainframe at list price would pretty much eliminate all the cost savings that arise from the lower mainframe software fees associated with less powerful mainframe processors. But IBM offers special discounts to mainframe shops that want to turn on the Linux engines hidden in their systems. IBM's best deal for shops buying new mainframes with less than 400 MIPS of native mainframe capacity is a $51,000 offering that includes a Linux engine and a copy of z/VM that lets the customer split the big Linux motor into a number of smaller virtualized Linux processors. Ordinarily, a mainframe Linux engine can cost $100,000 and z/VM for the engine would add $22,500 to the price.
For more substantial users IBM does even more to make Linux seem affordable. For example, users getting a new mainframe packing at least 400 MIPS can get a Linux engine for free. IBM will even toss in a copy of z/VM.
In either case, the user is still going to have to pay for hardware maintenance, which runs about $2,000 a month for that Linux engine.
And that's not the only Grinch at this Christmas.
Red Hat has joined in this deal by pricing its mainframe Linux distribution at $7,500, a big break from the $12,500 it usually charges when it's in a big IBM deal or the $15,000 a la carte price it cites on its web site. But after a year, the cost goes back to list price.
On top of the charges for getting the software, customers have to also pay for maintenance and support. z/VM maintenance for a single Linux engine runs roughly $3,000 to $6,000 a year; the price seems to vary quite a bit between deal packages. It seems to change with the season, too. Linux basic updating can cost $15,000 a year, but here, too, customers may be able to shop around and find slightly more attractive deals.
By contrast, in the X86 universe the snazziest Red Hat Linux, called Enterprise AS, costs only $2,500 a year for basic updates and support per system, which could mean a machine with two Intel or AMD processors. A server in this class packs about as much computing wallop as one of IBM's latest Linux engines, and if a customer thinks it doesn't, there are four-chip, eight-core AMD-based alternatives available. Comparable machines based on dual-core Intel chips ought to be widely available by the end of the year.
When you add in middleware, particularly DBMS software and its companion products, the financial contrast only becomes starker. Linux on a mainframe is a great idea, but even with all IBM's truly impressive giveaways, it's not a freebie. It's not even close.
IBM's Power chip users may pay a little more than shops with 32-bit X86 machines. Red Hat says it will sell and support Linux on IBM Unix Power boxes for the same price per machine that it charges X86 users. Novell/Suse prices p5 and i5 Linux at roughly the same levels, about $2,500 for two years, which is about $900 more than it charges for X86 boxes with up to eight 32-bit engines. The Novel/Suse Power price is the same as the charge for X86 boxes using 64-bit Itanium and Opteron chips.
At Red Hat, there's some fine print in the pricing, and users with complex distributions may find that pricing can increase by up to $1,000 a year for rich installations that need 7x24 support, but it's still dirt cheap compared to the cost of Linux on a mainframe. Users who just want Linux to provide a firewall or email services can stick with the standard offering; those who start putting serious applications into a Linux box generally go for the high end support schemes.
So, while there's sometimes a small gap between Linux pricing for basic X86 boxes and the cost of Linux on Power platforms, there's a big jump for mainframe users. The benefits of Linux living inside a mainframe's skin may be substantial, but the costs are not inconsequential. Nevertheless, the way mainframe shops work, installing an outboard X86 box or Power-based server alongside a mainframe isn't necessarily cheap, either. A lot depends on the skills pool at a particular site. For some shops, that mainframe Linux engine, which would look vastly overpriced to an outsider, might be the most practical alternative, or at least less of an error than it seems at first glance.
And then there's the political issue, which can be IBM's most effective pitch for big iron Linux. Mainframe computer folk view an enthusiastic X86 Linux team working alongside their proprietary team as a threat, and the better their results, the more likely it is that they make the legacy staffers nervous.
This is hardly a new idea. Leonardo da Vinci illustrated the need for caution in his wonderful fresco of The Last Supper, in the refectory of Santa Maria delle Grazie. It's been the most famous painting in Milan since something like 1495, and that's quite a distinction. Milan has plenty of nice stuff smeared on walls and, for those who don't worry as much about burglars as clerical Italian art collectors, on pieces of cloth, too. The Last Supper serves to remind us of the words of wisdom spoken at that event, "Separate checks, please."
Prospective users of Linux on mainframes and other proprietary servers might want to consider the recent adventures of one of our acquaintances. Bill knows a lot about computers and even has a prestigious college degree in the subject. He's not keen to waste a lot of time fooling around with computers, except maybe to make money.
Anyhow, Bill got a free 390 and decided to load Linux on it. The 390 in this case isn't a mainframe; it's a Thinkpad 390e made in 1999, and not even the snazziest model in the group. Still, using another machine, he downloaded the freebie Red Hat Linux, burned a set of CDs and fed them to the 390. The installation was the easiest he'd done since the days of DOS, he said, in an email written and dispatched using the Thinkpad.
He didn't just do this out of the blue. He found some support and advice on the web and, perhaps to his surprise, it turned out to be right on target. He also has learned that, as old as this computer might be, there's still a support section on the site that used to be IBM's and is now in transition to management by Lenovo. That support, the free microcode updates that allow the old Thinkpad to run the latest Linux or Windows, plus the abundant selection of memory, disks, optical drives, NICs, wireless NICs and spare parts make this bit of laptop history more viable than, for example, Multiprise 2000 mainframes built in the same year.
On August 17, we found a 2003-215 on eBay. (The link may be dead now because the seller bought only a 10-day presence and the clock was running out when we last looked.) It executes at 30 MIPS, or roughly the same raw power at the Thinkpad 390e with its 300 MHz Celeron, based on a 10:1 allowance to convert mainframe instruction rates to X86 clock speed. The mainframe has much more I/O bandwidth. It has 512 MB of main memory, twice as much as the 390e. It has about 250 GB of internal disk, depending on how the RAID is configured; the laptop has a mere 30 GB. The seller is asking $4,895 for the machine, less than 3 percent of its original cost, and it's been available for nearly 10 days with no takers at that price. There are a bunch of 390e machines on eBay. It's hard to find an exact match for Bill's 390e, but our best guess is that his machine, because it has upgraded memory and disk, would sell for at least $200, more than 10 percent of original cost.
So, if you think putting Linux on a mainframe or another proprietary box is a good idea because the more costly platform will have lasting value, think again. As Thomas Edison pointed out, the record speaks for itself.
— Hesh Wiener August 2005