The basis for HP’s November 1 Moonshot announcement, and in particular the Redstone low-energy server, was the realization at HP Labs more than five years ago that the real issue in server design is not performance but “delivering what the user needs,” says Partha Ranganathan,  HP Fellow in Nanostore Research and Development at HP Labs.

That soon led to the recognition that the developing hyper-scale computing users typified by very large Web services such as Yahoo!, Google, and Facebook have very different needs from more traditional large enterprises, he told Wikibon.org Co-Founder and Chief Analyst David Vellante and SiliconAngle Founder and CEO John Furrier in a live webcast interview from the announcement on SiliconAngle.tv.

These users do not need the fastest processors. They need the capability to do massive amounts of data movement in parallel. And to keep up with their huge growth they are buying servers by the thousands, which means that a $1 difference in price can equate to hundreds of thousands of dollars in extra expense. And it means that providing power, controlling heat, and conserving floor space are huge issues in their packed data centers. “The most important thing to look at is total cost of ownership, not raw performance.

“Five years ago we looked at all these processors designed for low power consumption for smartphones and starting thinking about how they could be used in servers,” Dr. Ranganathan said. But the issue extended beyond just building a new motherboard. It required new tools. And to get maximum value, the servers need to be designed as plug-and-play building blocks in a highly expandable federated architecture.

However, Redstone is only the first part of the total technology, which he says will literally turn IT on its head. “The basic computer has something that computes, something that stores, and something that communicates to other devices and end-users.”

HP Labs has been developing a very interesting technology for storage that it calls the “memister” – a memory transistor. Like traditional storage it is non-volatile but like memory it has nanosecond response, orders of magnitude faster than traditional storage. It can replace DRAM or flash storage and has interesting properties that HP Labs is still exploring.

Then for the next generation of networking, the third leg of the compute stool, “think about optics. Optics provides an energy-efficient, high-bandwidth channel. You put these three components together and you get a black swan event.”

Dr.  Ranganathan argues that combined with the demands of big data, this new architecture will replace today’s compute-centric IT design with a data-centric model to solve a growing problem. “Look at today’s servers. They have compute at the center surrounded by cache 1, cache 2, cache 3, and memory 1, memory 2, memory 3, a huge hierarchy. In 10 years we estimate that 80% of a server’s power will be used just in moving data around that hierarchy and other overhead.”

That is a disaster. To fix that he envisions what he calls the Nanostor built on memisters that will put all data at the center of the system with different kinds of processing optimized for the needs of different user groups surrounding that data. “So we start with Moonshot, add the other two legs, and we  have something very new. The data hierarchy will give way to a granular compute hierarchy.”

HP labs still faces challenges in realizing this vision, but, he says, “They can be managed.” And he predicts that the full data-centric architecture will be ready in this decade.


Watch live video from SiliconANGLE.com on Justin.tv