Getting a Grip on the zIIP
Published: March 7, 2006
by Hesh Wiener
In January, IBM said it would add the z9 Integrated Information Processor, or zIIP, to the roster of specialty engines in its flagship mainframes this year. For some workloads, a zIIP co-processor can reduce the amount of capacity required by DB2 under z/OS. This can cut the cost of software licensed under workload plans. No big iron shop can ignore a chance to get more computing for less money. But it's not yet obvious which shops will benefit from this option. It's up to each customer to decide for whom the bells and whistles will reduce tolls.
Basically, a zIIP is a mainframe engine loaded with special code that works with DB2 Version 8 or later and z/OS Version 1.6 or later. In principle, a zIIP should also work with Oracle or other database programs, but only if the DBMS software can work through z/OS to take advantage of the functions that a zIIP performs. Before long, IBM will publish documentation for the zIIP that will make it possible for all the software vendors and services companies serving large sites to use the zIIP. Until then, there will be some uncertainty about just how powerful, versatile, and friendly the zIIP will actually be. But even now, it's not too hard to sketch out some of the aspects of the zIIP that prospective users will incorporate into their planning.
For example, IBM has said that a zIIP will cost $125,000 in the United States. Prices elsewhere are likely to be similar. Maintenance, which will include firmware updates, has not yet been given a specific price, but in the case of the zAAP, a Java co-processor that closely resembled the zIIP in both technology and product positioning, IBM's maintenance charge is about $20,000 a year (or it will be from this coming April after a maintenance price hike). By contrast, a general purpose processor in a System z9, which uses the same hardware but has much different firmware, costs $800,000 to $1,200,000, depending on the kind of deal a user can wrangle, and maintenance costs in the vicinity of $100,000 a year. Clearly, if one or even a few zIIP processors can replace a general purpose engine, the zIIP provides some very substantial hardware and maintenance savings.
But while a zIIP is a piece of hardware, its real value is as a substitute for software. It is Hamburger Helper for DB2, stretching the work a particular DB2 configuration can perform at a much lower cost than mainframe meat could ever do.
So a user can try to figure out how much in software savings a zIIP might yield, and use that value, rather than the comparison to a hardware engine's cost, as a basis for justifying the co-processor. There are at least two crude ways to compare a zIIP with DB2 cost, and when more is known about the zIIP users will develop much more precise and accurate measures. Still, it's not to soon to try and make some rough estimates.
The narrowest measure of a zIIP's value might be based on the amount a user can save by reducing the computing capacity of a general purpose engine running DB2 by adding a zIIP. DB2 is frequently priced on workload basis. The more MSUs computing power units a DB2 partition consumes, the more it costs. Actually the formula is a little more complicated and is based on criteria described in detail by IBM's software pricing guide. But for a first cut, if DB2 usage is treated as a constant from month to month, the value of a zIIP might be compared to the number of MSUs the zIIP replaces times the cost per MSU of the DB2 workload based license.
For configurations with more than five general purpose engines devoted to DB2, the charge for the database software is at least $47 per MSU per month. Usually it's a lot more, because users don't install just the base system and instead add the Query Management Facility for another $24 per MSU per month, plus other enhancements.
One a three-year basis, that zIIP costs $125,000 plus $60,000 for maintenance or $185,000. That works out to about $5,140 a month or about the same as 70 to 75 MSUs worth of general purpose processor capacity. A 12-way z9 is charged out at 620 MSUs and an 11-way at 580 MSUs. So for a site with a z9 that uses about a dozen engines, the zIIP is actually more costly than a whole general purpose engine's DB2 in narrow software charging terms.
But DB2 also uses z/OS, and on a z9 in this power range, that can account for another $80 to $85 per month for each MSU dedicated to DB2 support, and that's for the base operating system alone. z/OS costs a lot more with the typical stack of extras. All of a sudden, the zIIP starts to look like more of a bargain, and this is without trying to apply hardware savings if a zIIP frees up any general purpose engines.
On smaller z9 configurations, a zIIP may pay off more quickly, because the cost per MSU per month of DB2 and z/OS is much higher for these less powerful mainframes. The way it is likely to work out, even a zIIP that's loafing might reduce software costs more than enough to pay for itself, and once again that's without any attribution for hardware cost savings.
This shouldn't come as a surprise to mainframe users, who know that DB2 on z/OS is a very expensive DBMS. Every shop spending big bucks on DB2 has probably looked into moving DB2 out from under z/OS, where the base price of DB2 Enterprise is about $25,000 plus maintenance. IBM offers a rich DB2 product for zLinux, AIX, Linux, Solaris, and Windows. Yet despite the price disparity, most mainframe shops running DB2 have, after looking, decided to stick with the premium software product.
Maybe migration isn't the reason IBM decided to offer the zIIP. Maybe there's another issue else that IBM is trying to address, a battle where the zIIP could be a very potent weapon. If, as I believe, it is not defection by users with growing established workloads, it is shops bringing up new workloads that view their options quite differently.
In addition to the obvious pure mainframe alternative, new database applications mean new candidates. There are platform choices, such as Unix, Linux, and Windows. For each platform, there are DBMS alternatives. Of course there is Oracle, with which IBM can contend, Sybase, and the various open source, SQL-enabled databases that have earned increasing respect from corporate users. But the product that makes IBM most nervous is Microsoft's SQL Server, not only because it offers attractive value for money, but because it requires Windows, the environment that led IBM to frustration and ultimately defeat on the desktop, the dominant server environment today, and the environment that seems to be in contention for new projects at many big iron shops.
And that suggests that the big question for DB2 users, and of course for IBM's mainframe hardware and software teams, is this: Can a zIIP enable a mainframe running DB2 to compete with SQL Server? We'll see, when IBM delivers the zIIP.
IBM Previews zIIP DB2-Assist Mainframe Engines