Oracle's core problem
"We don't have a position with respect to dual-core processors. A core is equal to a CPU, and all cores are required to be licensed. Therefore, if you have a dual-core processor, you are required to have two processor licenses." -- Oracle, 2005
"At the fourth Open Source Performant Database conference in San Jose, 2007's star start-up MetaPeta announced that more than 250 Fortune 500 companies – and nine of the top ten – were now clients. "We are delighted with the rapid acceptance of our product and services," said Jason Mitchell, the 29 year old chief executive of MetaPeta. "The figures speak for themselves". He celebrated the announcement during his keynote speech by having his four year old son Tux double the performance of an MPGrid XV system through plugging in five more PlayStation 4 consoles. "He can pay for the software licences out of his pocket money", Michell told the crowd, "and still have change for a Larry Ellison inaction figure."" -- ZDNet newsfeed, 2010.
It's not likely that General Electric or Texaco will be running their businesses on games consoles any time soon, but it is absolutely guaranteed that the next five years will see servers make a complete transition to multicore processors. Nothing will stop this, because it's now the only practical way to make faster chips.
It is also true that the most economic and effective way to make multicore chips will be to go as massively multicore as the basic technologies allow. While Intel and AMD are gingerly feeling their way into that world with their existing designs, IBM has taken the opportunity of a clean slate to introduce the Cell chip with nine cores and promised performance in the stratosphere. It's going into games consoles, it has to be cheap. The bang per buck will be astonishing.
Which brings us back to Oracle's problem. If it insists on sticking to its arbitrary per-core licence -- why not license per GHz, or by amount of storage attached? -- then it will become not just the major expense in a database, but overwhelmingly, massively so. Even a relatively low-power database program will give Oracle a run for its money if it can run in five hundred processor cores for the same price as Oracle running in thirty two. Needless to say, open source has the ideal licensing regime for multicore: fudging the TCO figures to say otherwise will prove a challenge to the most inventive marketing team.
Oracle's current strategy will be disastrous. If the company insists on sticking to it, then we can only hope the death throes won't slow down the rest of the industry for too long.