The DB2 Heresy
Published: April 29, 2008
by Hesh Wiener
IBM mainframe shops that use a relational DBMS prefer DB2 under z/OS, but it's not their only option. They can run a more generic version of DB2 under z/Linux or under Linux or AIX on Power, or under Windows on X64 iron. They can even run yet another DB2 variant on Power-based iron that is embedded inside i5/OS. In fact, users willing to think outside the z box can find plenty of DB2 servers that won't be suggested by mainframe peddlers; some of them can be cheaper than an all-z solution. In the glass cathedral, this idea is a heresy. Yet it can be a valuable heresy, the way an Amdahl coffee mug (or mainframe) used to be.
IBM no longer has to compete in the mainframe processor market, and if its lawyers crush Platform Solutions, chances are it never will. Even so, it is folding its mainframe server manufacturing business into its Power Systems group more or less the way it has consolidated its proprietary midrange line with its Unix variant business. And it never stops tinkering with its scheme of what amounts to processor pricing based on how the engine is used, designating specially priced CPU cores as IFLs, zIIPs, or zAAPs.
IBM faces no competition in the mainframe operating systems field, unless you can make a case that z-compatible Linux from Red Hat and Novell are alternatives to z/OS, z/VM, or z/VSE. Most observers treat z/Linux as an additional environment for IBM mainframe users, not one that customers can move to if they feel that z/OS might not be their best choice. It's just not practical for z/OS shops to move their legacy software to z/Linux; it's all that most of them can to just to keep from falling too far behind as IBM pushes z/OS updates out the door.
IBM faces real and respectable competition in middleware, particularly when it comes to DBMS packages and their surrounding constellations of related programs. In the relational DBMS segment, even with its advantages in the z/OS base, IBM is a distant second to Oracle, a financial fact that always seems surprising to mainframe shops running DB2. IBM could very easily slip to third place any minute now, overtaken by Microsoft. This is apparent from the of the RDBMS market prepared by Gartner and IDC.
Once you look outside of middleware and examine the applications universe, IBM is not even in the running. IBM hasn't built an ERP business around DB2 the way Oracle has surrounded its DBMS with products that add value. (It used to, two decades ago, but sold those businesses off to become the Switzerland for systems.) Nor has it developed (or bought) a suite of banking applications, insurance industry programs, or big utility billing packages, even though IBM mainframes claim a lot of hardware and operating systems turf in these segments. The main thing IBM likes to buy, it seems, is its own shares. Well, even though in some ways it looks like IBM is hollowing out its mainframe business, Wall Street remains in love with the company. (Of course Wall Street liked mortgage-based derivatives, too.)
Most application suites that work with DB2 under z/OS can be adjusted to work with IBM's DB2 for Linux, Unix, and Windows, which is the product that IBM ultimately has to pump up if it wants to fight back against Microsoft and pick up large enterprise business that might otherwise go to Oracle.
Even though IBM does not love talking about its more generic version of DB2 when it is selling software to z/OS shops, it does have excellent libraries of support documents for all its DB2 products. One of the key IBM documents is a guide to the differences between the SQL dialects used by its z/OS DB2 and DB2 Universal Data Base products. The document is about a thousand pages long and growing, but that's not because it takes so much space to discuss only the differences between its two flavors of DB2. The cross reference is actually a very fine reference manual spanning the DB2 universe. Its size, structure, and content provide substantial evidence that IBM wants DB2 to gain ground in all the software environments where it is supported, even in the Windows world where IBM's prices (and Oracle's prices, too) are far higher than the fees Microsoft charges for its SQL Server products.
To compete, IBM has to price its DB2 for Linux, Unix, and Windows at an attractive level. And, under IBM's imaginative and somewhat confusing Value Unit pricing scheme, the pricing pattern IBM offers for DB2 running on Intel or AMD chips, where Microsoft is beginning to tear holes in IBM's glass house franchise, is echoed on Unix and Linux boxes, including not only servers made by Hewlett-Packard, Sun Microsystems, and Dell, but also IBM's own System p Power machines and the z/Linux IFL engines inside IBM System z mainframes.
Basically, IBM rates a z9 IFL at 100 Value Units for pricing purposes, and the list price for the core Linux DB2 Enterprise package comes in at $386 per value unit, or a one-time charge of $38,600 including the first year's maintenance. Actually, IBM always seems to be cutting deals and offering discounts. A submission for benchmark testing of DB2 on Sun hardware (based on a server that has, like a z9 IFL or Power 5 server, a rating of 100 Value Units) put the base product cost at roughly $28,000, which works out to a 25 percent break from nominal list price. A for-core Power5 server (called a 560Q) with a 1.8 GHz engine that could easily run DB2 costs $43,800 on IBM's Web site. If a Power5 core running DB2 delivers performance comparable to an IFL, which is rated at the same number of Value Units, then a single core devoted to DB2 would cost less than $11,000, a lot less than an IFL, which can cost as much as $125,000. If you run DB2 on all four cores and you can get the price Sun put in its TPC-H report, the basic DB2 cost comes to $112,000 and adding in the server you end up paying about $156,000 (plus the cost of other DB2 components). A single IFL and 100 VU of DB2 might come to $153,000, or about the same amount of money except if you go for the Power box you might get four times the DB2 performance for the same money. (The computations for a Power 6 or z10 engine are essentially the same, but those processors are rated at 120 VU per core rather than 100.)
There is no easy way to compare the power of a Sun engine to a Power engine, although you might scour the libraries of benchmarking organizations to size things up. But what is clear is that the fast Sun engines, pegged at 100 Value Units, are unlikely to be only half as fast as a Power 5 when running DB2 and might be quite peppy. If they are half the speed of a Power platform, Sun plus DB2 might still offer twice the price/performance of an IFL. If DB2 worked well on the Sun Niagara Sparc T1, T2, and now the T2+ chips, which have eight cores and either four or eight threads per core, the match gets trickier. IBM rates a Niagara at 30 Value Units per core or 240 VU for a whole chip. We haven't spotted any published standard benchmark tests that have a Niagara running DB2, but if the combination works at all we are confident it would give an IFL a run for the money.
If you want to go all the way to X64 country, where CPU cores are rated at 50 Value Units (possibly because they have half the power of a p5, but possibly because IBM wants to make DB2 cheaper to compete with SQL Server). IBM offers Intel Xeon and Advanced Micro Devices Opteron servers with four or eight cores. These machines cost even less than Power boxes with similar muscle, and the numbers show that an IBM System x with DB2 might be the best deal in town.
Of course the whole idea of using non-z DB2 instead of the z/OS version on a z machine only makes financial sense if mainframe DB2 is more expensive than z/Linux DB2. Well, one way to keep costs down is to run z/OS DB2 with the help of a zIIP, because that can reduce the size of the software's MSU footprint and at the same time eliminate the cost associated with more MIPS running in a standard mainframe engine using z/OS and other software associated with the DBMS function. Unfortunately, we can't spoon feed you the price of a DB2 under z/OS setup that eats up general purpose engines and zIIPs. You have to look at your own software bills to figure it out, or, if you think you can adjust your system and save some money, a proposal from an IBM rep or reseller for the hardware and software you'll need to run your workload under z/OS at the lowest price IBM can come up with. But even without any other information, one thing is pretty clear: A Power box with DB2 costs less than a zIIP and might do four times the work. . . and it would also remove the (reduced) DB2 footprint from a z box, which has to save some money.
The point is that even if DB2 for z/OS was free you might have a smaller DBMS budget buying a Power (or Sun or HP or X64) platform and putting the Linux, Unix, and Windows version of DB2 on it.
None of this addresses the technical aspects of the matter. It might be that the hassle and cost of moving legacy DB2 work to the other DB2 UDB version rules it out. But even if you can't move old work, if you are thinking about new workloads for DB2, it would be foolish to dismiss the possibility of using alternative platforms including an IFL, a Power box, an X64 server, or even a non-IBM machine. At the very least, a little bit of heretical talk with your IBM rep or reseller might help you get improved conditions if you ultimately decide to preserve your current z/OS hardware and DBMS.
Post this story to del.icio.us
Post this story to Digg
Post this story to Slashdot