Sure, you want computing power, but how much are you willing to pay for it?
That's the question posed by Tycoon, a Hewlett-Packard program that would weigh the priorities of groups jockeying for a shared computing resource. The computer maker believes the best way to set such priorities is to let customers bid for the resources they need -- whether with real money or through another budgeting method.
HP initially envisions the mechanism as a way to allocate resources with a company. But eventually, it could lead to companies bidding for processing power delivered by a third party or pooled from their own machines.
Tycoon is part of HP research geared toward making its Adaptive Enterprise concept a reality. That concept calls for a company's computing infrastructure that can quickly respond to changing business needs, but HP believes real-world firms are too complex to figure out in advance how to allocate computing resources.
"We do not try to imagine every possible trade-off," said Russ Daniels, chief technology of software and Adaptive Enterprise. "That problem is not tractable."
Thus HP developed Tycoon, which sets priorities through a marketplace rather than centralised planning. "If I bid more money, I get more resources," said Kevin Lai, an HP Labs researcher.
HP showed off the technology to reporters here Tuesday as a way to set priorities for digital animation jobs. But the technology is escaping HP Labs: Tycoon could be used at an animated film festival in the United Kingdom called SE3D, and it will be used at CERN, the European Organisation for Nuclear Research, which operates a large cluster of computers shared by scientists processing data from particle collision experiments.
Tycoon tackles a thorny problem at the heart of not just HP's Adaptive Enterprise, but also competing schemes such as IBM's on-demand computing and Sun Microsystems' N1. Those initiatives seek to make computing infrastructure more efficient by sharing multiple tasks on the same systems instead of having multiple systems dedicated to separate tasks.
But sharing can be difficult -- especially for "server-huggers" reluctant to hand control over their own machines to central computer administrators.
Diving into the pool
HP is testing a new program called SASU -- Shared Application Server Utility -- that pools application servers that run Java programs. The service lets administrators perform an extensive analysis of server usage patterns to squeeze more applications onto the same server. That lets HP save money on application server software that's paid for by the processor.
"We have a couple patents on algorithms to do this resource allocation," said Kumar Goswami, manager of HP Labs' utility infrastructure management department. "We've taken things running in silos and put them in shared hardware."
The algorithms also work to allocate tasks across groups of blade servers -- thin systems that slide into a chassis with shared power supplies and network switches -- he added.
The sharing initiatives -- broadly grouped under a concept generally called utility computing -- haven't been easy to create, however. HP cancelled a product called Utility Data Centre for lack of customer interest, though some of its ideas carry on in blade servers, management software and other technology.
