Guild Companies

Despite all the noise that people (including myself) made about supporting Linux on Intel, proprietary, and Unix servers, people are still trying to figure out what to do with Linux. And they are doing so in a near vacuum of information about how other companies are using Linux on various platforms. But IBM, which is pushing the most Linux revenues right now, and Hewlett-Packard, which is pushing the most Linux server shipments, both have a very good idea of what is going on.

It's hard to say for sure, but if the economy stays bad, if Windows 2003 is expensive to implement or is buggy, and if Microsoft's Licensing 6.0 pricing meets with the lukewarm acceptance in the midrange that I expect it will, these pressures will push companies that do not want to learn another platform to give Linux some serious consideration, whether they like it or not. IBM knows this, and that's why it is trying to position itself as the powerhouse for Linux sales in the data center. HP knows this, and that is why it is trying to do the same thing.

According to statistics compiled by Gartner's Dataquest research unit, IBM captured 41.6 percent of the $385 million in Linux server sales in the U.S. market alone last year. IBM hasn't seen market share statistics like that in the server market--especially in a new market with lots of aggressive players and on an operating system platform that it does not control--since the 1970s. (Part of the reason has been the adoption of Linux on the zSeries and iSeries, which results in a lot more money for a server or partition sale than for Linux on an Intel-based server.) Whether IBM can hold that share is debatable, especially as Linux becomes more pervasive, but in the U.S. market IBM's share of Linux server sales is accelerating. Moreover, both Hewlett-Packard and Dell are each, with about 65,000 Linux servers shipped in 2002, outshipping IBM by a factor of two to one. As Linux takes off--among small and midsized business looking to save some dough and midrange shops frustrated with Windows--HP and Dell could accelerate their sales faster than IBM does.

The fact remains, however, that IBM is positioning Linux as an alternative to Windows and Unix for budget-conscious customers. IBM knows that everyone is trying to figure out how to consolidate their servers and get the utilization of the machines up above the typical 10 to 15 percent while not adversely impacting the performance of their applications. Linux can run in a logical partition on the iSeries, and Windows NT, Windows 2000, and Windows 2003 cannot. There is not a server consolidation play on the iSeries when it comes to Windows. While the Integrated xSeries Server (IxS) coprocessor for the iSeries and the Integrated xSeries Adapter (IxA), for attaching outboard xSeries servers to the iSeries, is a great solution for hybrid OS/400-Windows computing, the IxS and IxA setups only allow storage consolidation on the iSeries disk arrays. The IxS processors and IxA-attached outboard machines run just as inefficiently as a stand-alone Windows server ever did. And you can just forget about dynamic allocation of processor, memory, or I/O resources for Windows. That's just one reason why Linux is a better option as an adjunct operating system on iSeries servers.

Linux is, however, an alien operating system to a lot of people, and that is a serious barrier for system administrators and programmers to jump over. I will remind you that people said the same thing about Unix in the 1980s, and however strange Unix was, when compared to MVS, VSE, VM, CPF, and SSP, the IBM mainframe and midrange operating systems, and other proprietary counterparts at IBM's competitors, Unix eventually came to represent half of aggregate server sales a year--a position that it still holds. If the disparate Unix environment can take the server market by storm--winning half the market outright and winning the other half by weaving itself into proprietary operating systems--then Linux, which is in many ways a single, cheap, standard Unix that has yet to grow up with SMP scalability and enterprise-class reliability features, can eventually secure a similarly impressive piece of the server market. People will and are jumping into the alien territory that is Linux. (I'm going to load Red Hat Linux 8 on my Dell workstation as soon as I am done writing this essay. And I'm using a fresh disk drive, so as to not completely screw up my Windows setup.)

According to IBM's internal market assessments of the worldwide Linux opportunity, there is not yet a lot of action in Linux when it comes to OLTP, ERP, batch, and CRM workloads that are common on AS/400 and iSeries servers. The data is so thin that, in 2003, it hardly registers. So when you hear that Linux usage here is doubling, tripling, or quadrupling between 2003 and 2006, take it with a grain of salt. The data I've seen says that IBM isn't expecting these workloads to add up to even a quarter of a billion dollars by 2006. Similarly, Linux platforms for application development, which account for a few tens of millions of dollars today, will hit about $300 million in 2006, according to IBM's market stats, and data warehousing and analysis servers will be lucky to hit $400 million by the same time. However, supercomputing and clustering workloads running on Linux servers are expected to grow from about $500 million in 2003 to over $1 billion by 2006; e-mail and workgroup processing will more or less double, to about $700 million, over the same term. The infrastructure workloads--where the Linux has been over the past few years--will account for about $3 billion in server sales, about double the sales for 2003.

Based on the activity in the press, in the market statistics, and in people's attitudes about the economy, I think the shape of this data is valid, but the scale of the numbers could be a lot higher. The growth curves could start to bend upward more sharply if people think they can get better value for money--whether it is true or not--from Linux. More important, Linux is as much a cultural phenomenon as a technology, and the perception is that if you build skills in Linux, you are building a future, marketable skill. Ditto for Windows. It's tougher to make that case for any variant of Unix, and harder still for OS/400 or the many-named MVS.

In a survey across companies in the Western economies, IBM found about 65 percent of customers using Linux for Web serving, with another 50 percent using it for network serving, another 50 percent using it for Web appliances, and nearly another 50 percent using it for firewalls. Yet another 45 percent use it for application development, and another 40 percent use it for e-mail serving. A minority of customers are using it for e-commerce, workgroup, technical applications, data warehousing, or other things. (These numbers add up to more than 100 percent, because a Linux server can be used for multiple tasks, obviously.)

Here's the real interesting bit of data I have seen, and it concerns the poster child for Linux on the iSeries: server consolidation. So when companies consolidate off platforms and move workloads onto Linux servers, just where are they coming from? Well, according to IBM's stats, 60 percent of the platforms that get consolidated onto Linux are coming from Windows NT server environments, and just under 30 percent are coming from Windows 2000 environments. For print and file serving, Linux plus Samba (the Windows print and file serving clone) works about as reliably (or, in some cases, crankily) as Windows itself. I say this from personal experience. My home network is based on a FreeBSD server running Samba, which has some trouble with my HP photo printer because the drivers are terrible, but, generally, it works and I don't even know I am not on a Windows server. A very large proportion of companies moving to Linux is coming from Windows, according to IBM's research. Small wonder Bill Gates is nervous about Linux. But I am cynical enough about the computer business to know that if Gates didn't have a real enemy out there in the market, he would invent one to get all of these lawsuits off Microsoft's back. IBM did much the same to wiggle out of the Consent Decree in the early 1990s, when it was down on its luck and the Unix server market was hammering the mainframe and AS/400 monopolies throughout the decade. There are no clone mainframe or midrange memory makers to speak of any more, and the clone mainframe makers (Amdahl and Hitachi) are out of the biz, and there never has been a clone AS/400 maker. These are not accidents.

Back to Linux server consolidations. Just under 50 percent of platforms that consolidate to Linux are coming from Unix environments, and this makes perfect sense, since Linux bears a strong resemblance to Unix. Moving to commodity Lintel platforms is the easiest jump that entry Unix server shops can do to save a lot of money, and as Linux matures midrange shops will be able to make a similar jump.

Another 10 percent of platforms consolidated to Linux are mainframes, and another five percent are coming from AS/400 and iSeries platforms. Another 10 percent or so come from other, more obscure platforms, such as NetWare. (Again, all of these numbers add up to more than 100 percent because companies are consolidating multiple platforms down to Linux machines.)

This may make you laugh if you don't know anything about the iSeries, but the virtual LAN, virtual I/O, dynamic logical partitioning, and other sophisticated technologies that are part and parcel of the modern iSeries midrange machine make it a very good platform on which to consolidate workloads into a Linux environment. Linux plays with OS/400 on an iSeries in a way that Windows does not and cannot, until a native version of that operating system is available for the Power family of RISC processors that are at the heart of the iSeries. (That will not be coming any time soon, despite all the rumors. Windows on Power is like an Elvis sighting. Until he plays you a song and you step on his blue suede shoes, don't mention it to your friends. That said, I think Windows on Power is a great idea, and now that there is a 64-bit version of Windows--Windows 2003, due April 29--it certainly is possible.) With real 64-bit computing on the iSeries, with more partitions per machine (10 partitions per Power4 processor on the new iSeries line), a strong economic case can be made for the iSeries as a Linux consolidation platform. And when and if IBM can get the number of logical partitions on an iSeries box up to 255 in total by the time the Power5-based machines ship next year, an even stronger case will be possible. But before any of us can do anything useful with Linux, we have to find out what it is and how it works.

IBM may have blazed the trail for virtual Linux partitions on the iSeries, but it is by no means that far ahead of the pack. Similar technology is available for Linux partitions on IBM's pSeries Power4 and Power5 servers, and the more sophisticated partitioning from the iSeries is being ported to the pSeries. You will also be able to make a case in the near future for Linux partitions for the "Pinnacles" Itanium 2 servers from HP. Unisys is supporting Linux partitions on the 16-way and 32-way ES7000s, which support IA-32 and IA-64 processors, with partner SCO Group now, and even Sun Microsystems can run Linux within its dynamic domains (and some customers are in fact doing this, although this is not a supported product from Sun at this time). And with Microsoft acquiring Connectix a few weeks ago, Windows servers will soon be able to run Microsoft-endorsed Linux partitions; VMware already offers Linux virtual machines on Wintel and Lintel iron, and SW-Soft offers virtual Linux slices on a Linux box and will soon support virtual slices on Windows boxes.

The real question is who is using Linux in commercial production, and while IBM's internal data doesn't necessarily answer that question, Judy Chavis, worldwide director for Linux at HP, sheds some light on the situation as far as HP sees it. HP has been a lot more quiet about its Linux activities than IBM, but it has not been slouching by any stretch of the imagination. She says that HP booked $2 billion in Linux-related sales across hardware (including printers, servers, workstations, switches, and such), software, and services lines in 2002. It's hard to say for sure because of the complexities of the HP-Compaq merger, but she reckons that in 2001 the two companies collectively had about $1 billion in Linux-related sales. Here's the interesting bit. She thinks that HP can maintain its roughly 30 percent market share in Linux servers going forward (that's about HP's share of sales in the past 20 quarters according to Chavis), riding up the wave in Linux sales. At the same time, she believes that an expanding software and services--particularly services--component in Linux-related sales can drive HP to double those sales--and maybe even more--in 2003. Right now, most of HP's Linux sales come from hardware, but in the 2003 to 2004 timeframe, she thinks (and hopes) that services will comprise the lion's share without a drop off in hardware sales. Eventually, perhaps two-thirds of HP's Linux sales a few years from now will come from services.

As for customers, the perceptions are that most commercial Linux customers are using Linux to displace Unix-based supercomputers or Windows-based infrastructure servers is not, as far as HP sees in its accounts, entirely accurate. Chavis says HP sees very little of this going on, but adds that HP is happy to pursue this business if customers want to do it. She says that HP's Linux server sales break down roughly into thirds these days: one third to Web infrastructure, one third to HPC, and one third to hard-core commercial data processing at manufacturers, retailers, distributors, and financial services companies. This is Linux running real, database-driven workloads. When pressed on the cannibalization issue--Linux eating into Unix sales--she says that there is some replacement of HP-UX servers by Linux machines, but claims that Sun is much more exposed because of its heavy dependence on infrastructure workloads. Apparently ERP applications running on HP-UX and AIX are more immune to porting and pricing pressures than infrastructure workloads, which is why Linux doesn't play so well there. And the pricing pressure on entry servers, which are typically used to support Web infrastructure, were never big markets for HP and IBM as they were--and continue to be--for Sun. But it is reasonable to expect that as Linux matures, it absolutely will be something that hard-core Unix shops will examine for application servers and maybe even database servers if they think they can save some money.