Dell is positioning itself to supply systems for cloud builders, including enterprises that want to construct private clouds.

So far, Dell may be the name best recognized for designing a handful of consumer and business systems, then relying on a tightly controlled supply chain between the customer's online order placement and Dell's delivery of the product.

One the things that Dell has mastered is online order capture and customer product configuration, skills that have served it well in the business and consumer mass market. In dealing with large cloud suppliers, or "whales," as such customers are known internally at Dell, there may be only one order and one configuration -- but the delivery could be of thousands of matching systems.

One such "whale" that Dell supplies is Facebook. Facebook uses thousands of servers to maintain millions of consumers' Facebook pages, which are constantly moving around pictures and text. Dell is also the main supplier to the Microsoft Azure cloud, according to Barton George, Dell's chief cloud strategist, who dropped into the Cloud Computing Conference & Expo in Santa Clara Tuesday.

George agreed, in an interview with InformationWeek that if Dell becomes a major supplier to cloud computing, it will be a new chapter for the online retailer. Among other things, the company must draw lessons from what large cloud suppliers prefer in a server, then offer a few choices of similar servers to smaller cloud providers, known internally as "dolphins."

There are only a few whales - builders of massive Internet data centers - such as E-Bay, Google, Yahoo, Facebook and Microsoft. But there could be many dolphins.

There's no requirement that clouds be built of x86 instruction set computers, with processors from AMD or Intel, and some clouds are being built on IBM Power or Sun Microsystems UltraSparc processors. But the economics of cloud computing dictate that clouds be built with low-cost, reliable parts, and that they be easily managed and easily virtualized. The broadest set of tools and skills to do meet the latter two are found today in the realm of x86 servers.

Dell is finding, George said, that the speed of assembly and price of hardware are not the only factors to which a would-be cloud supplier must pay attention.

Dell has typically sold a range of models that it designs from what it deems the most cost effective parts available at a given time. Business buyers are often looking for high reliability so they order servers with redundant disks, power supplies, fans or network interface cards, ensuring that a business-critical application will keep running even if one of the components fails.

Cloud computing architects, on the other hand, have a different set of interests. They often do not like redundant parts, even though hardware failures represent an unwanted interruption that must be coped with in the cloud as elsewhere.

Working with a diagram on scratch paper, George illustrated how cloud developers kick the problem of a hardware component failure upstairs into the software running the cloud. The model dictates that the workload of a failed machine be moved elsewhere and picked up where it left off, or rerun. The failed server eventually gets repaired or replaced.

George quoted Dell's distinguished engineer and chief architect for cloud servers, Jimmy Pike, who has pointed out that "the resiliency is in the software, not the hardware," so Dell must adjust to that concept when designing cloud servers. Dell has had some practice. Microsoft's Azure cloud is to be announced in two weeks at its Professional Developers Conference in L.A., but Dell has been designing servers for Azure for three years.

So one of the first elements in cloud design is to strip out redundant parts, lowering the cost per server. A second element is to build in strong virtualization features, which include lots of memory, multi-core processors and, frequently, multiple host bus adapters for storage traffic or multiple network interface cards for network traffic, if the virtualization workload is likely to have lots of I/O.

There is no one design that is right for every cloud, George said, but at the same time clouds, including private clouds, will have many requirements in common and Dell must master the best practices of producing the right machine for a given cloud.

To help customers build private clouds, Dell's Data Center Solutions Group will offer templates of proven cloud servers and let customers browse among them. It will also cite best practices in the software stacks that go on those servers to become nodes in a cloud. Dell's DCS group already has experience in the field, George said.

"The way we look at it, cloud is not binary. It's not all or nothing in any one direction. It will be mix of IT (traditional IT infrastructure, with mainframes, Unix servers), a virtualized private cloud (based on x86 servers), and public cloud," he said. In other words, there will be collaboration between the data center of the future, which will include a private cloud, and the public cloud, such as EC2.

Some at the Cloud Computing Conference called such a federation the "hybrid cloud." George doesn't care what you call it. He just wants Dell to provide the basic building blocks that will run it.