Big Blue Clarifies Software Pricing for Multicore Machines
Published: April 10, 2006
by Timothy Prickett Morgan
When is a processor core not a core? When your software vendor says so, that's when. The advent of processors with multiple cores has made a mess of software pricing, and chip makers, operating system providers, systems software suppliers, and application software vendors all have a different take on how they think software should be priced in this multicore world. Now, it is IBM's turn to "clarify" its position on software pricing.
The newsy bit of this announcement is the fact that IBM, like Oracle, is being generous in regard to the multicore "Niagara" Sparc T1 processor from Sun Microsystems. The funny bit is how IBM twisted the English language around to justify its pricing policy.
Rather than just come out and say "not all cores are created equal," IBM used what it called the "vendor definition of a 1-way server" to justify its position, and then tweaked its software pricing accordingly.
"Multicore chips that primarily reduce power and heat may not generate significant additional system performance when compared to the same system with a single identical core," IBM said in the announcement. "In these cases, IBM recognizes that you are not able to derive the same additional throughput as you could from a system with the same number of single core chips."
IBM then explained in the announcement that for IBM machines based on Power processors, a core was the vendor-defined 1-way, and that for IBM software (such as operating systems, middleware, or databases), each of those Power4 and Power5 cores in the pSeries, iSeries, and OpenPower machines, the PowerPC processors used in the BladeCenter JS21 servers, and the pricing would be one core requires one license--except for the JS21 blades and the OpenPower machines, which would only require one license for every two cores. (Consistency is everything.) In the mainframe world, where Linux can be installed on mainframe processors, a core is a 1-way, and it requires a single license.
For the HP 9000 server line using PA-RISC 8700, 8800, and 8900 processors, a core is a vendor-defined 1-way equates to a socket, and the PA-8700 has one core and a PA-8800 and PA-8900 chips have two cores. But, it takes two licenses for this vendor-defined 1-way--see how useful this term is? And yes, you should be laughing about right now--not one license as the definition of a 1-way would seem to imply. IBM's 1-way is twice as good as HP's, at least as far as software is concerned, and it is four times as good when it comes to the OpenPower machines and the BladeCenter JS21 servers. For HP Integrity servers, a socket is a core is a single license, but it would have been too difficult to just say it that way, apparently.
In the realm of the Sparc, IBM concedes that Sun calls the dual-core UltraSparc-IV and UltraSparc-IV+ processors a vendor-defined 1-way (well, Sun says a socket is a socket, no matter how many cores are in it, and software pricing should be based on a socket), but doesn't cut Sun any slack on this for these chips. If you want to run IBM software on these chips, you pay for two licenses for each socket--twice as much per vendor-defined 1-way as IBM is charging for its own dual-core Power4 and Power5 chips and four times as much as it is charging for the dual-core PowerPC 970MP processors. (That's fair, right?) Following in the wake of Oracle, which is charging two licenses for an eight-core Sparc T1 processor (which does about the same amount of work as a two-socket, Xeon server using single-core chips), IBM has decided to cut Sun some slack. For the Sparc T1 chips that have four or six cores activated, IBM will charge two licenses, and for a fully enabled Sparc T1 with eight cores running, IBM will require three licenses.
Now, if you are talking about 32-bit or 64-bit Intel Xeon or AMD Opteron processors, you get a real great deal. In this case, a vendor-defined 1-way is a single chip, and whether that chip has one or two cores, you need only buy one license for this.
To recap: If you buy IBM's most advanced, dual-core Power5 or Power5+ processors and you want to run IBM software on them, you will be penalized. If you buy HP servers and all but the Sparc T1 machines from Sun, you will be similarly penalized. But if you opt for Intel or AMD processors with the same or slightly lower performance, you will cut your software bill in half. And your systems will be a lot cheaper, too.
Maybe IBM's Systems and Technology Group, which designs Big Blue's servers, and Software Group, which creates its operating systems, databases, and middleware, ought to rethink this one. Imagine if IBM said software costs would be at parity for all Power-based machines, not just the dual-core PowerPC 970MP chips. Imagine if IBM demonstrated superior performance and software pricing at the same time.
But, having already clarified this, IBM will probably not further clarify its pricing and thereby give itself more advantages than it is giving to competitive X64 chip architectures. By doing it this way, IBM has surely calculated that it can extract more software profits from the market, and that's what really matters, right?