HP’s Project Moonshot Aims to Recreate Servers, Again
In the late 1990s, there was a shift in thinking around how servers could be made and how several of them are designed to share space. The idea was to pack several server computers in the space that had previously been required for just one by making certain parts smaller, eliminating others and sharing resources like power and cooling in a single assembly.
Now we call them blade servers, and today they account for about 15 percent of the world’s servers, with vendors as varied as Hewlett-Packard, IBM, Oracle and Fujitsu.
But the fundamental problems facing those who buy servers in large quantities remain the same: Technology demands more computing cycles, which servers with ever more powerful chips can certainly deliver, but companies have limited space to put them, limited power resources to run them and cool them, and limited ability to pay for it all.
Today Hewlett-Packard aims to change the discussion about the future of servers with something it calls Project Moonshot. The idea is pretty straightforward: Cram 2,800 servers into a single rack that would today house a few dozen, or at most 128, blade servers. Make them all share the same internal networking, cooling and power supplies and generally boost the number of servers that can fit into a defined space. One way or another, more efficiency is badly needed, and as Parthasarathy Ranganathan, a Fellow at HP Research I talked to yesterday, told me, the time has come to stop trying to squeeze “blood from a stone” in order to get it, but rather do something more radical.
The headline that everyone is paying attention to is that HP has selected an ARM-based chip from a Texas-based start-up called Calxeda as the chip it will use in its development platform, called Redstone. ARM, as you know, is a flavor of chip technology designed by the British firm ARM Holdings that’s widely used in mobile phones because it is very power efficient. ARM has recently started to make some inroads into general-use personal computing against the Intel- and AMD-based world of x86 computing.
But it’s less important to focus on the chips that HP is using here than on the fundamental shift that HP is trying to create. “We’re not trying to start a new chip war,” Glenn Keels, HP’s director of marketing for HP’s Hyperscale business, told me. There’s no reason that Intel’s Atom chips couldn’t one day be just as suitable for this. Make no mistake, though: ARM chips are coming to servers, one way or another.
Aside from the Redstone development platform, HP also aims to let potential customers kick the tires of the Redstone-style servers by running their applications on them and seeing how they perform versus traditional servers in a series of development labs that the company will open around the world. The first will be in Houston, and it will open in January. More will follow next year in Europe and Asia.
HP also named a handful of partners that are participating in the Moonshot project by contributing hardware, software and technical expertise. Among them are AMD; Canonical, the company behind the Ubuntu flavor of Linux; and Red Hat, the enterprise Linux company. It all looks very interesting, and if HP can nudge the industry in a direction where millions of servers packed into data centers can consume significantly less energy than they do now, everyone should be happier. The benefits would be the increased availability of computing power at a lower cost, with less relative energy consumption and therefore less impact on the environment. It’s hard to argue with any of that.
Here’s an HP video explaining what Project Moonshot is all about.
(Image is the cover of an extremely inaccurate 1959 children’s book imagining what a routine flight to the moon might be like for a 6-year-old boy.)