The announcement was something of a non-event, since HP was only saying that it would use the Centerton chip, it didn't announce any systems or configurations. That will come later in the year when Centerton is in full production. The timing of the event was surely intended to send signals that HP/Intel relationship was alive and well, even as HP, along with lots of Intel execs as witnesses, meet Oracle in court over what's now sure to be the ill-fated Itanium chip and HP's Integrity servers.
As the sordid spectacle of that trial unfolds, a picture is painted of an unhappy Intel that didn't want to continue losing money on the development of Itanium after an ecosystem for the chip failed to materialize. Indeed, HP is about the only user of the chip now, and had to change its relationship with the chipmaker so that Intel was more of a contractor that would produce the chips as long as HP footed the bill for R&D and production. Oracle released dozens of internal HP emails, presumably gotten during discovery, that illustrated the tension over Itanium.
[ What is the opposite of a microserver? Read IBM's Sequoia Is World's Fastest Supercomputer. ]
At the press conference this week, we were assured that with the Centerton announcement, the companies are back to their old collaborative selves. Centerton is a member of the Atom family of system-on-a-chip (SOC) products, but it has certain server-class features which previous Atom chips don't. Key features that make it appropriate for microserver use are: a 64 bit architecture that is now common for Atom chips, virtualization support in hardware, ECC memory support, dual-core design, and an extremely low power profile. The Centerton chip draws just 6 watts. Compare that with top of the line Xeon chips that can consume north of 130 watts, albeit with many more cores running.
The market for microservers was demonstrated by SeaMicro, which uses low power Xeon and Atom chips to pack lots of servers into a small space. The company claims its systems can take just 1/6 the space and 1/4 of the power of commonly used racked servers from other vendors. AMD purchased SeaMicro for $334 million in February, but the company has yet to release products based on AMD chips.
HP's claims at its press conference go well beyond what SeaMirco has done. HP speaks of thousands of systems per rack, with space savings of up to 94%, which works to about a 20-to-1 ratio of its microservers to standard 1U systems.
The burning question with microservers is the market size. HP thinks the new systems will satisfy up to 10% of the server market, while AMD pegs the number at 20%. Clearly the systems are intended for "scale-out" environments, where services and applications can meet demand simply by deploying more and more servers running the same software. While there are certainly still many "scale-up" applications that run as single instance and need more memory and faster CPUs to meet an increased demand, but that sort of application design is now largely out of favor. Whether you come from a SOA background or a Web services one, it's been drilled into the architectural mindset to keep services tight and easily replicated. That notion matches up well with microserver designs, so one could imagine such systems being used for everything from Hadoop and MapReduce to Web services and systems to even properly architected enterprise applications.
At this year's InformationWeek 500 Conference C-level execs will gather to discuss how they're rewriting the old IT rulebook and accelerating business execution. At the St. Regis Monarch Beach, Dana Point, Calif., Sept. 9-11.