IBM has made much of the performance of its new Power6 chip. What it hasn't shouted about is how the new chip has affected its software pricing -- and instead, has created a mess that looks set to confuse users. The only clue has been a letter to users.
At the processor's launch, IBM's product manager Ross Mauri said that the Power6 beat all its competitors in standard benchmarks, while returning twice the performance of the Power5 processor for the same energy consumption. It also claimed that the chip is now the fastest-running Unix processor -- IBM makes variants that run as fast as 4.7GHz.
So far so good. But it's in the processor's performance that the problems lie -- not from a technology perspective but because of what it's done to IBM's price list.
All software vendors are struggling to find an equitable way of pricing products in a world where the terms 'socket' and 'processor' have lost their meaning. So in July 2006, Big Blue launched a new method of software pricing that it said took account of how different chips perform. It invented the PVU or processor value unit. Rather than being priced on a per-socket or per-processor basis, you pay for your software according to the functionality and performance extracted from the hardware and software combination.
This all sounds fair enough until you start to look at the detail. To arrive at a PVU for each processor type, IBM has to assign to it a value that reflects its performance. It does this using benchmarks -- but how it decides which benchmarks to run and how the systems are configured is known only to IBM. As anyone with benchmarking experience knows, there's no such thing as a perfect or universal benchmark, only one that measures the performance of a given system running under a particular set of circumstances.
It might appear that the system is intended to make the relationship between functionality, performance and price equitable to both parties. In practice, IBM doesn't want you to pay too little for its software and get more performance out of its software than before simply because chip vendors now sell processors with multiple cores.
So in fact, the new pricing model serves mainly to obfuscate. The main reason for this is that the metrics used to calculate the PVU are not open. You can't ensure that you're getting good value for money, and you can't compare price performance across platforms.
That's because the system doesn't just concern IBM's own Power chips, as IBM's software also runs on CPUs from Intel, Sun, HP and AMD. So a comparison of the PVUs attached to different processors is revealing. The Power6 is king of the mountain at 120 PVUs per core, while IBM's Power5 and System z, Sun Sparc IV, and Intel Itanium are priced at 100 PVUs per core. Intel and AMD's x86 server chips languish at 50 PVUs while the UltraSparc T1 squats at the bottom of IBM's performance pyramid at 30.
How IBM arrived at this formula is unknown but users haven't been slow to denigrate it. According to one commentator on a story in The Register, IBM's formula suggests that the best value for money is obtained by buying an x86, rather than an IBM Power6.
"DB2 is thread sensitive and you can buy 2.2 x86 threads for the same price as each Power 6 thread you purchase, essentially getting over 1.5 times as much real world work done," he says.
Another response suggests however that the Power6's parallel processing capability makes this an unfair comparison, while one points out: "The term you are looking for is Confusopoly which is becoming a standard response by large corporations when the product they sell is a commodity with no discernable differences."
IBM's not alone in this type of software pricing: Oracle uses a processor metric akin to IBM's in pricing its higher end products.
What's clear is that this kind of pricing is inequitable because it's not transparent: it's based on benchmark results, which can be neither verified nor repeated in the real world. As one commentator points out: "For fairness, the derivation of the core performance factor should be transparent and validated by a third party. IBM, as a processor and systems vendor, as well as a software vendor, has the power to manipulate software licenses [sic] to benefit their [sic] own servers."
The history of the software market suggests that users are intolerant of apparently unfair and opaque pricing. If IBM can't clarify the basis on which it allocates PVUs, it may find the whole exercise counter-productive in the long run.