Oracle has quietly jacked up software prices on IBM's Power6 iron by 33 per cent, after removing a multicore scaling factor that was in effect to reduce prices.
The change was spotted by an intrepid reader at El Reg, and can be confirmed with a visit to Oracle's website.
After coming under pressure in late 2004, for pricing its database software as if a single core were the same thing as a single-socket processor, (in the days before multicore processors became mainstream, and in anticipation of the day when they would be,) Oracle launched a new pricing scheme.
It allowed customers using multicore chips to multiply the processor chip count by a factor of 0.75 and use that number, rather than the number of cores, to calculate per-processor software license fees. Before multicore chips, a socket was a processor, so it was easy. Five months later, after much complaining on the part of server and chip makers that not all processor cores are created equally in terms of performance, Oracle introduced a different set of scaling factors for a wider range of processors. To be fair to Sun Microsystems, Oracle gave the "Niagara" family of multicore chips a 0.25 scaling factor, dual-core x64 processors were given a 0.50 scaling factor, and RISC and Itanium chips received a 0.75 scaling factor.
Over the past four years, of course, chip makers have introduced more powerful processors, some with more cores, some with simultaneous multithreading, others with faster clock speeds, and some with a mix of these to boost performance. The latest Oracle processor core scaling factor table, which you can see here, now rates a Power6 processor at a scaling factor of 1.0 per core, which means no more 25 per cent discount on Power6 processor cores when it comes to Oracle products (most importantly, on database software).
The Power6 processor, launched initially in the summer of 2007 and rolled out in the Power Systems server line last April, is a dual-core chip, just like its predecessors the Power5 and the Power5+ kicker. But the Power6 chip has roughly twice the performance of those initial Power5 chips in online transaction processing workloads. So you can understand Oracle's temptation to remove the discount core scaling factor from the machines and put them in the same price band as IBM's System z mainframes.
There is some speculation that the change has to do with the multiple shared processor pool feature of the Power6 machines, which allows a group of logical partitions on a single Power Systems machine to be capped with a certain number of processors with the idea that you try to drive up utilization on the cores in the pool. If customers are cutting back on cores allocated to Oracle databases and getting the same performance, that is a kind of price cut, too. There is also some confusion about pricing in virtual or logical partitions versus virtual private servers, such as Sun's Solaris containers. The confusion is cleared up in this Oracle document, which clearly states that "soft partitioning is not permitted as a means to determine or limit the number of software licenses required for any given server."
The Oracle price hike on Power6 chips seems unfair given that the quad-core Sparc64 VII processors used in Sun and Fujitsu machines have the same 0.75 scaling factor, but you can bet Sun and Fujitsu are happy about this because it makes Oracle software cheaper on their big iron. Ditto for quad-core x64 chips from Intel and Advanced Micro Devices and dual-core Itanium 9100 series chips from Intel, which currently have a 0.50 scaling factor. (That is effectively the same thing as saying a Power6 core is worth two x64 or Itanium cores.)
It will be interesting to see what Oracle does when Sun and Fujitsu roll out the Sparc64 VII+ quad-core chips, which they are expected to do soon, and when Intel gets its quad-core "Tukwila" Itaniums out the door in June or July. The Tukwila chips should certainly get their scaling factor removed, but given that HP and Intel do not have a database software business, I would venture that Tukwilas might sneak by with a 0.75 scaling factor.
To be fair, Oracle should run a database benchmark test on each processor and come up with a literal scaling factor based on possible clock speeds of all processors and make the scale all relative to the performance of one machine that it picks as the gold standard. But software companies want something that still looks and smells like per-socket pricing and is easy for their salespeople to peddle.
In Oracle's defense, IBM's analog to their processor scaling factor methodology, which it calls value unit pricing and introduced in the summer of 2006 after watching the Oracle experiment, is a similarly loose scaling metric for processor-based pricing. It charges more for Power6 processors for IBM's own DB2 and other systems software than it does for Power5 systems, as you can see here. Power6 and System z10 mainframes have a rating of 120 per core, compared to 100 for Power5 cores and 80 for Power6 chips used in the JS12 and JS22 blade servers sold by IBM. (That 80 rating is due to the limited memory and I/O bandwidth on the blades, as far as I can tell.) IBM rates the initial Sun Niagara chips at 30 VUPs per core, and the second generation at 50 VUPs per core, the same as dual-core, quad-core, and hex-core x64 chips. Itanium, various other Sparc, and other chips have a 100 VUP rating. It is probably a fair guess that Intel's Tukwila chips will get a 150 to 200 VUP rating when they come out, and it is hard to imagine a 16-core "Rock" UltraSparc-RK chip not being in the same range when it comes out sometime in the fall.
If IBM doesn't buy Sun and spikes the Rocks, that is. ®