Intel has heavily virtualized its servers in two areas: in the data center consolidation of legacy applications and in its shared electronic design automation tools. Both sets of virtualized servers are candidates to become different types of internal clouds, he said.
But first, enterprise cloud builders have to match what they're about to do with the user expectations of what such a cloud would provide. The appeal of cloud computing is that it supplies "the illusion of infinite resources," because only a handful of users will be trying to exploit large amounts of capacity at any given time. In the normal course of events, when they need multiple servers, they will tend to be available, he noted.
But even if user expectations are matched by what the enterprise is willing to build, "There is no off-the-shelf configuration of servers and server interconnects" for cloud computing. It may sound simple to assemble a large number of x86 servers, and Amazon.com, Google, Microsoft and Yahoo "have done it very well," Waxman continued. But they have put substantial engineering effort into the process. There's no blueprint to consult or cloud building packages to purchase. Other private cloud builders are going to have to make the investment as well.
At the same time, there is incentive to try. The existing cloud data centers are known to be cost-effective forms of computing. Amazon charges outside users 8.5 cents per hour for a Linux server in EC2. Servers amount to 50% of the expense of data center outlays, with electrical power and cooling for servers another 23%, said Waxman, according to careful studies of IT budget outlays.
Clouds work and achieve economies of scale because "they optimize (hardware and software) on many different levels," he pointed out. Not only are servers more fully utilized as hosts to multiple virtual machines, but the software running in the cloud is optimized for its function there. Search companies optimize their search engines to run effectively on clusters of x86 servers, but even within an identifiable industry segment, "search platforms use different server optimizations for the way each goes about search."
In the future, x86 servers will be designed with fewer components and fewer components in upright positions, impeding the flow of cooling air as it passes over the circuit board. Racks of servers will share a power supply and cooling fan, and all the parts will be allowed to run hotter without damaging components, he predicted. Intel is shooting for components that can run continuously at 40 degrees Centigrade or 104 Fahrenheit temperatures.
Improving voltage regulators on servers, allowing a two-way server to scale back to 60 watts when idling, would result in a savings of $6 million a year in electrical costs for a company running 50,000 servers, he said.
Building private clouds with servers that use flash drives instead of disks will contribute to cooler operation and 90% less electrical consumption that disk drives, he added.
But the private cloud assembly will also need to have a high degree of manageability as well if it is to maintain its economies of operation. Failing to put the right elements together may defeat user expectations of large amounts of resources at low prices, making the private cloud an undertaking with inherent risks, he concluded.
InformationWeek and Dr. Dobb's have published an in-depth report on how Web application development is moving to online platforms. Download the report here (registration required).