Mad Dog 21/21: It's i or Die for Power in the Midrange
Published: March 1, 2010
by Hesh Wiener
Intel has a very strong server chip lineup. Its current Xeon line, dubbed Nehalem, will soon give way to two newer product lines, Westmere-EP and Nehalem-EX. Where performance counts more than anything, Power still trumps X64. But where economy is foremost, X64 usually wins. With its new chips, Intel is ready to take on Power in the midrange. IBM's server business cannot live on mainframe-class machines alone. Consequently, IBM may be about to lose its grip on servers to systems built to shared standards, much the way it lost its power in PCs.
Advanced Micro Devices will also get its next round of CPUs into the market later this month, the 12-core Opteron 6100s, and Intel will not have much of an edge in the X64 market with its six-core Westmere-EPs or its eight-core Nehalem-EX chips. Both vendors will soon have very rich Xeon and Opteron lines. The 32 nanometer Westmere chips are like 45 nanometer Nehalems but have two more cores in the same thermals and offer some architectural enhancements. Servers with four or more sockets will move from current six-core Xeon 7400s (based on the old Intel front side bus architecture) to Xeon Nehalem-EXs with the new QuickPath Interconnect. IBM is bound to use these new chips in its own x Series servers.
Once AMD and Intel have both begun shipping new circuits, the battle for market share will be fierce. Power chips will be caught in the crossfire. (So, too, will Sparc.) The proprietary chips don't enjoy the huge volume advantage of X64 CPUs. They don't support software markets that have the comparatively huge opportunities provided by Windows or Linux on X64. So how can IBM win business for its Power chip family?
Users who are deeply committed to i5/OS and i 6.1 are likely to stick with what they know, at least for their legacy workloads. Users currently running AIX selected this environment after considering alternatives. Linux might seem like less of a leap for AIX shops than it would be for, say, z/OS devotees, but it apparently doesn't look that way to users. Linux is running on a much larger percentage of mainframes than on Power Systems machines. So unless circumstances peculiar to a particular shop damage the relationship between IBM and the customer, Power is safe where it already lives and Power usage will grow at least as fast as legacy applications, and faster if the customer favors IBM environments for new applications, too.
But, as I've said, X64 systems generally have a price advantage, and it can be dramatic when the comparisons are at the low-end of Power power, which is the midrange for X64 hardware. The potential savings can tempt customers and, because there are so many Power shops also running Windows or Linux on X64, this temptation is not theoretical; it's a daily issue for IBM's Power marketing personnel. And it's not a new phenomenon, but rather a newly more serious one.
Last summer, researchers at Ideas International pointed out that the first round of Nehalems offered per-core performance that was great enough to challenge IBM's Power6 chips. While IBM subsequently came out with midlife kickers for the Power6 generation (doubling up the processors in a box, but not in the chips themselves) and, just a few weeks ago, noticeably faster eight-core Power7 chips, IBM has not moved down into the high volume end of the server market, thus ceding that turf to products in its own x line and the alternatives sold by Hewlett-Packard, Dell, and others--including IBM's own System x and BladeCenter products.
The result is an echo of IBM's mainframe history in the i and AIX machines, one that is unnecessary and potentially very harmful to IBM and its midrange server customers.
Here's what i and AIX shops have to worry about:
When IBM decided that it was no longer interested in building small mainframes it lost three important parts of its commercial computing ecosystem: wannabe mainframe shops that could only afford incubator systems, small ISVs whose budgets often make the tiniest mainframe shops look wealthy, and universities that trained computer science students to use mainframes.
After pulling out of the small mainframe business, IBM supported Flex-ES mainframe emulators sold by Fundamental Software. The Flex-ES lines even included mainframes in laptops that were used by quite a few software developers and a number of IBMers, too. But IBM later decided to curtail licensing software for (and some patents used in) Flex systems, that that was the end. Oh, there is said to be a very lively underground where users illegally run IBM software on Hercules mainframe emulators, and some software developers may be living in that largely invisible world, but commercial users don't run production software on Hercules systems.
IBM is aware of the costs of its decisions and has tried to recover some of its losses. It has made a big effort to court educational institutions, and apparently made considerable headway. It seems to have become willing to look the other way in some cases where Hercules emulators are used by ISVs (and perhaps IBMers, too) for software development. But even so IBM has had a hard time selling mainframes, which lately it has blamed on the impending end of its z10 product cycle.
So far, IBM hasn't applied any lessons from its mainframe experience to the Power Unix market, where a programmer or student cannot find a workstation that runs AIX. Small user organizations are similarly priced out of entry. This wouldn't be such a bad thing if it were not for the availability of cheap Windows, Linux, and Solaris workstations. (HP offers workstations except for its HP-UX Unix variant, making it somewhat more likely that its Itanic product line will sink, leaving the company with exactly the same range of computers that Dell has.)
Users of i boxes who want a cheap development platform cannot get a workstation any more than their AIX cousins can. The only option for i or AIX Power developers who want a cheap platform is getting lucky on eBay or exploring the used computer market. IBM does offer entry-level servers in its Power lines, but IBM's idea of entry level is costly compared to entry level X64 iron. The ante might not discourage developers who have concrete plans, but there is no getting around the fact that pretty much every computer science student in the world and, come to think of it, just about every university student with any major either owns an X64 computer or has ready access to one.
IBM is not struggling to get new customers for its Power platforms except by picking off established users running Sun boxes and finding unhappy HP shops. That is not the way to make its business thrive, and it is not how it must act if it wants its Power platform base to be optimistic rather than defensive.
This crabbing about IBM's distain for little guys probably doesn't register with IBM. In Armonk, it looks like the same old story they've been following for years, perhaps decades. They know the trend started when the typewriters no longer sat on so many desks and ended when nothing in the typical office client and local server cluster had an IBM nametag. However, IBMers note, Big Blue is bigger and richer than ever, or so it seems.
Well, the IBMers are in error. HP is bigger than IBM, and livelier, too.
IBM's strength is that it has figure out how to be the biggest player in computer services, which in practice turns out to be the business of running a hospice for legacy mainframe applications. But even that business could lose its sparkle, and if it does the lower-margin Power offerings may no longer be viable.
This time around the story doesn't necessarily end with IBM slaying all the dragons. This time, X64 hardware can do the most popular tricks that used to be reserved for other platforms, and particularly the proprietary Power servers sold by IBM.
One big factor is virtualization. While most Intel server chips sold during the past few years have had some hardware support for virtualization, the technology really became pretty good with Nehalem thanks to the huge improvements in the memory subsystem and interconnect. It's going to become even better with Nehalem-EX and Westmere-EP. VMware keeps learning and the X64 virtualization technologies from other software makers, including Microsoft, are getting better all the time, too. This is serious: Microsoft even supports virtualization for ordinary end users running 64-bit Windows 7 on old the Intel Core Duo processors. Virtualization is how Microsoft gives Win 7 boxes with loads of legacy code a way to preserve applications that predate Win32 technology; these boxes are pretty common in corporate offices. Whether or not that have the best virtualization software, the folks in Redmond have made everyone in computing aware that virtualization does in fact work on X64.
Virtualization is at the heart of a marketing issue that has now become an important strategic issue for customers. It used to be easy for IBM to argue that a Power box with virtualization is the efficient alternative to racks of disparate X64 servers; now that is a red herring. The X64 servers that are coming to market this year are going to make a self-evident case for X64 consolidation on X64.
IBM won't have a very easy time persuading mainframe shops that zLinux is the only way to reduce heat, clutter, and complexity in a glass house. IBM won't be able to lure many X64 Linux users to Linux-on-Power. And in the i market, IBM may find that its virtualization technology will be popular among users who want to eliminate i footprints but not very effective when it comes to pulling X64 jobs into an i system.
IBM just might have to rethink its pricing and show real interest in entry Power customers. If IBM can't do this within the financial constraints imposed by Power chips, it will have to rethink its strategy.
In the midrange, IBM's last best hope may be i rather than AIX, because i drags along proprietary software, a unique culture, and ease of use that really appeals to people who basically just need bookkeeping systems. The catch--for IBM and for users, too--is that there's no dirt cheap entry system. IBM doesn't offer a way to go from zero to a point where an i on Power system is big enough to demonstrate its potential value. Those very small users are getting picked off by X64 products nearly every time.
IBM's i executives should start thinking the unthinkable: Maybe IBM ought to offer an i-on-X64 starter product line.
Brace Yourself For A Bevy Of Server Chip Announcements
Post this story to del.icio.us
Post this story to Digg
Post this story to Slashdot