The Four Hundred

Only a small fraction of the OS/400 installed base can scrape up the dough to get to the semi-annual COMMON midrange user group and expo, which was held last week in Orlando, Florida. We were there, and the top brass at IBM made the case for their iSeries strategies and gave an update on how biz is going. Gartner analyst Tom Bittman also delivered a keynote speech to size up on-demand computing and where the iSeries fits into this world.

Bittman actually got to talk first, but iSeries General Manager Al Zollar had the real top billing, since he is not only watching the iSeries but also steering its course through the rough waters of the server business. "Through the first half of the year, the iSeries has been fantastic," Zollar told nearly 2,000 people who attended the iSeries Nation Town Hall meeting on Sunday afternoon. Revenues for the iSeries line for the first six months of 2003 were up 12 percent from the same period last year. Shipments of iSeries servers were also up 12 percent during this time, and Zollar went on to say that shipments were up 15 percent in the second quarter of 2003. Zollar said that sales in the Americas region were up 20 percent, with 24 percent growth in the United States. Sales in the Asia/Pacific region were up 20 percent as well. This means the European market is seriously dragging down the averages. Europe was the first market to widely accept Unix back in the late 1980s and early 1990s, and something might be going on here again. Very likely, the smaller businesses in many European countries are figuring out that they can do all of their work on mirrored Wintel and Lintel boxes.

While Zollar was able to honestly claim that the iSeries had gained market share in revenue--according to IDC, aggregate revenues for servers were down 3.3 percent in the first quarter and only up two-tenths of a percent in the second quarter; shipment growth for the iSeries is nonetheless behind for the average across the entire server market and significantly behind that of the Wintel and Lintel markets, which I reckon are growing at two to three times the rate of the iSeries shipment increase. Nonetheless, that iSeries shipments and revenues are up by double digits is a big improvement. The revamping of the iSeries line in January has gone a long way toward improving sales, but so has a pent-up demand for server capacity as long-delayed projects are finally getting some funding.

Here was an interesting statistic that Zollar tossed off in his COMMON presentation: The amount of aggregate iSeries computing power shipping in the first half of 2003 is up a staggering 40 percent. You heard that right. Shipments are up 12 percent, revenues are up 12 percent, but aggregate capacity is up 40 percent. This seems to suggest that some midrange and high-end OS/400 shops are buying big boxes and consolidating many smaller servers onto big iron. It also suggests that these companies are getting their iSeries processing capacity for a lot less money and are buying a lot of extra capacity now, which they expect to use later. (In an aside, Zollar said that more than 700 processors were shipped under capacity-on demand contracts in the second quarter of 2003, and that, to date, some 4,000 iSeries machines have been shipped with capacity-on demand under the skins.)

Under a very simple financial model that I built using the numbers Zollar talked about at COMMON, I can show that IBM can increase overall shipments in 2003 by 12 percent (to 31,000 or 32,000 servers) and can increase revenues by 12 percent (to about $1.9 billion) by shipping average boxes that have 25 percent more capacity in them than they had in 2002 and by cutting the cost per CPW unit of capacity by 20 percent at the same time. I looked at this table for a while, and I just keep wondering what happens in 2004 if the active iSeries customers today are buying so much excess capacity? What is going to make them buy next year, or the year after? I saw this happen in 1998 before the Y2K crisis. I hope even more IT projects will come online, and IBM is clearly hoping this, too. To be fair, IBM's job is to make money now and to worry about the future spending plans of OS/400 shops when it gets there. My job is to think about the future from outside of the Big Blue box, which is why I am bringing this up.

In that spirit, I'd also like to know how many AS/400 and iSeries footprints were lost to data center consolidations and server consolidations in the first half of 2003. My guess is that it is at least several thousand. How many have been lost to competitive replacements? That could be in the thousands, too. How many workloads that could have been on OS/400--like application servers or Web servers--have been shifted to Wintel, Lintel, or Unix platforms? It's hard to say.

These questions have me a little worried. Zollar said that in the past few years, IBM has been adding anywhere from 3,000 to 5,000 new customers to the OS/400 platform each year, and he said further that his team is committed to hitting 5,000 new customers in 2003. But if customers are unplugging footprints faster than IBM is adding them, this will not make independent software vendors believe that the OS/400 platform is a good market to go after. I can remember, in the late 1990s, when the new customer rate for the AS/400 was on the order of 12,000 new customers a year and IBM shipped 50,000 or 60,000 boxes a year. I have a hard time getting excited by new customer and worldwide shipment numbers that are half this level, much less a revenue stream that is half the $3.5 billion to $4 billion IBM used to bring in by selling AS/400 iron. (When you add in storage and operating system revenue to the comparisons, which used to be separate revenue streams for the AS/400, the comparisons are even less flattering.)

What I am worried about is that IBM is turning the iSeries market into a mirror image of the zSeries market. IBM doesn't really sell puppy mainframes any more, and it doesn't seem concerned about that. It's like IBM is conceding the low-end and midrange parts of the market to other platforms. I think it is unwise and unhealthy for IBM to do the same for the iSeries, and while the new iSeries Model 800 and Model 810 machines are great compared to past AS/400 and iSeries machines, Xeon-based Wintel and Lintel servers will beat them hands down on many modern workloads, including IBM's own beloved, much-touted, and under-used WebSphere. This is an untenable position for long-term growth for the iSeries business. I have some suggestions for how to fix this, and I will start laying them out next week.

Tom Bittman is the lead analyst at Gartner for a number of server platforms, including the iSeries and Windows. Many years ago, he was a software development manager at IBM Rochester, before he moved over to Gartner to become one of its top analysts and most frequently sought after analysts, particularly for public speaking events like the COMMON keynote address.

The keynote that Bittman gave on what Gartner calls the Real Time Enterprise was interesting, but it mirrors many things we have heard from all the major IT players, including Gartner at its IT symposium only a few months ago in San Diego.

To sum up Bittman's presentation: IT organizations are too inflexible, and it takes too much time to implement and adapt solutions to changing conditions in the marketplace. Even large amounts of money can't make some IT projects succeed. IT organizations are slow to react to change, and IT infrastructures have simultaneously become more complex, which further complicates matters.

Another big problem in IT is the inefficient use of resources, explained Bittman. Somewhere between 60 and 70 percent of the aggregate computing power that customers have installed is unused. Improvements in workload management and logical partitions have helped matters, but the virtualization that these and other advances entail are not useful unless they can be dynamically allocated and reallocated, and, most significantly, unless their administration can be automated, so workloads get enough capacity based on the business needs and service-level agreements that reflect those needs.

The Real Time Enterprise, which is akin to IBM's broad eLiza/On Demand initiatives from the past several years, is about virtualizing and automating the IT infrastructure that companies use to support applications. "It doesn't matter if you have five nines on a box," says Bittman. "You have to have five nines on a service." This is certainly true, but it has taken years to get even that high of a level of availability on a box, to put it bluntly. Unreliable boxes do not make for reliable services. And he also knows that only recently have server capacity and software prices fallen enough that small and midsized companies can even think of clustering or moving Intel-based workloads onto logical partitions on much more expensive machines. It takes time to get the components of a service reliable, and it will take time for companies to think about delivering services, rather than supporting specific applications on specific boxes. "We don't see the Real Time Enterprise as a step function," he says. "This is a long-term process. Real Time is a path, and we're all on it."

Bittman also echoed one of the things I have been saying for quite some time. As logical partitioning and other virtualization technologies increase the use of resources, companies are going to buy fewer resources. Couple that with the improvements in technologies and the cut-throat competition between suppliers of servers, storage, operating systems, and middleware, and there is a tremendous downward pressure on revenue growth in the server market. Bittman seems to think that just because vendors know this, they will be able to charge enough extra for automation services and virtualization features to make up the difference and still be able to grow server revenues at a modest rate between now and the end of the decade. I have my doubts about this. I think that server revenues could start trending downward as virtualization really takes off. It's hard to say at this early state of the game. Unix, Windows, and Linux are just starting to get viable virtualization technologies, and they represent the vast majority of the server market by shipments and revenues. We'll just have to wait and see.