HP Debuts Utility Computing Services

by Timothy Prickett Morgan

Server maker and utility computing pioneer Hewlett-Packard last week rolled out the next iteration of products involving utility computing, a computing model it has been pushing under various names for about six years now. Having tried a number of different approaches and go-to-market strategies, HP is doing something that looks a bit like what rival Sun Microsystems is attempting to do with its Compute Grid and Storage Grid offerings, but with its own twist.

HP has a number of very high-flying utility computing customers already, notably film maker DreamWorks, which uses HP's in-house utility to help it do some of the rendering for its movies. Such utility computing deals are designed mainly to allow customers to buy on-demand computing capacity to handle peak workloads, so they only have to keep enough capacity on site to cover normal workloads. This can significantly reduce the amount of money that companies spend on digital media rendering, oil and gas exploration, bioinformatics, simulation, and other numerically intensive workloads. This might fall short of HP's original lofty goal with the Utility Data Center (UDC) offering from November 1999, which proposed to completely virtualize the server, operating systems, storage, and networking infrastructure of the data center and allow companies to buy capacity as they needed to run their workloads. This was a big concept, and even as it was demonstrated as a working product in 2001 and improved in 2003, HP discovered that it was also a very tough sell. And then Sun bought Terraspring, the company that invented the software at the heart of the UDC offering, and last September, after trying to get traction with UDC, HP broke up the UDC offering into a more modular pieces and figured out how to weave its own new utility computing software (from the Consera and Novadigm acquisitions) into a more flexible offering.

Rather than announce utility computing products--the UDC was a box running software that virtualized your data center--last week's announcements are services created by HP and based on iron owned and run by HP in its own data centers that it sells to customers under the marketing umbrella of Flexible Computing Services.

Sun, through its Compute Grid and Storage Grid, and IBM, through its On Demand Supercomputing Centers, are already offering similar utility services. Sun is trying to help its customers--presumably to be called utility service providers some day--build the utilities that in turn will sell capacity to end user customers, HP is more focused on offering services that allow customers to buy capacity on infrastructure that HP manages. In effect, this will be a productization of the DreamWorks deal and other similar deals HP had for customers who wanted the benefits of a product like UDC, but who did not want to install and learn it, not matter what the benefits to infrastructure flexibility, utilization, and return on investment. HP has undoubtedly figured out that, being one of the largest corporations in the world with a sophisticated IT infrastructure and a large consulting business with wide and deep knowledge that HP would probably do better selling utility computing on its own boxes than selling utility computing as a product to go into other people's data centers. That doesn't mean HP will not offer a wide spectrum of utility computing offerings--including capacity on demand and metered capacity on its servers as well as custom-built utility data centers for clients. HP will, to put it bluntly, sell customers whatever they want to pay for, and with utility computing still being a relatively new and untried technology after six years, the company has to do this because the market could move in a number of different directions as people become more comfortable with the idea of utility computing.

While HP has been offering utility computing as a private service on a one-to-one basis, either building utilities at customer's data centers or in its own as a private service, Brian Fowler, utility services global director at HP, says that it has yet to do a one-to-many, public service that is based solely on using HP's own data centers to support many different customers and changing workloads.

Right now, Flexible Computing Services has two services, one called IPS and the other called IPS+. With the IPS service, which is short for Infrastructure Provisioning Service, HP has set up servers, operating systems, and all of its utility computing gizmos and middleware--including elements of the old UDC solution as well as other components that Fowler will not specify--in its data centers in Palo Alto, California; Houston, Texas; Littleton, Massachusetts; Paris, France; and Bristol, England. The specific box count was not reveled, but HP has installed a mix of 32-bit Xeon, 64-bit Xeon and Opteron, and 64-bit Itanium servers and can provision Linux and Windows on the Xeon and Opteron iron and Linux, Windows, and HP-UX on the Itanium iron. HP could offer OpenVMS, its proprietary operating system from the DEC server line, on the service, and because Solaris 10 can run on HP Xeon and Opteron iron, it could offer Solaris Unix support as well. But Fowler says that customers have not requested the latter yet, so they are not part of the offering.

With the IPS service, you pay for computing capacity based on the number of hours you use a particular type of processor and its operating system; a 32-bit Xeon processor running Linux has a list price of about 55 cents per CPU per hour, with an Itanium CPU running HP-UX costing about $1.50 per hour; 64-bit CPUs based on Xeon or Opteron platforms running Windows and Linux cost somewhere in between. Customers can install any applications they have licenses for on these machines and use them as they will, but obviously at these prices, it does not make much sense to use it as a Web server unless you want to compare the cost of managing your own servers, fully burdening it with people costs in the data center. The IPS+ service adds workload management, batch scheduling, and grid software (specialized clustering software for supercomputing clusters) on top of the raw IPS offering, which makes it suitable as an extension to an existing internal compute grid or, if you don't already have one and don't care to, the compute grid you have no desire to own but wouldn't mind renting from someone else. The IPS+ service allows customers to license or rent these software tools. HP's initial partners for the IPS+ offering include Platform Computing, DataSynapse, United Devices, Altair Engineering, and PathScale.

These two services, IPS and IPS+, are available today, and they are the basic utility computing offering that the market expects and maybe--just maybe--a bunch of companies will pay for.

Further down the road, HP will roll out targeted applications on a utility basis, and is starting with a service called Application Provisioning Service, or APS, some time in the first half of 2006. This builds a particular application set, targeted at specific industry verticals, on top of the IPS+ offering. The first vertical that HP will attack is computer-aided engineering, and it is partnering with Abaqus and MSC Software for structural analysis, Fluent for computational fluid dynamics, and LSTC for finite element analysis. All of these vendors have agreed to license their software on a pay-per-use basis, but they are still finalizing their pricing.

To help stir up interest in utility computing and to get some more experience dealing with a wide variety of end user companies, HP has established a Flexible Computing Club that gives customers startup and consulting services and 48 CPU hours worth computing capacity on 32-bit Lintel iron for $5,000 to do a pilot project. If customers end up putting their workloads into production on a utility contract, HP will give them a rebate on the $5,000.