For big power savings, HP thinks smaller
November 2, 2011
HP announced yesterday that it is testing chips from an Austin, Texas mobile processor company for use in Hewlett-Packard servers. Calexda is one of a host of chipmakers who produce ARM processors. HP said it believes these chips can provide the needed horsepower for server tasks -- while sipping power, instead of gulping it like anything from Itanium to Xeon chips.
HP is very serious about reducing power consumption at its enterprise customer sites. The vendor has a road show in play that addresses this benefit of moving off older HP 3000 hardware. The PCI-based N-Class and A-Class servers reduced power consumption (as measured by BTUs) by 30 percent over the 9x9 Series. And the Integrity 2660 class of servers, similar to an A-Class, use 567 watts at idle to support an entire server.
A mobile chip solution for enterprise establishes a fresh measuring tape for power usage. HP is calling the initiative to create a new server line Project Moonshot. It hopes to start selling these ARM-based servers by next year. The rollout at the Calexda HQ in Austin showed off the EnergyCore ARM system-on-chip (SoC) for cloud servers and on-demand processing.
Hewlett-Packard isn't planning to introduce hardware for its small to midsize customers anytime soon that utilizes EnergyCore ARM. The math on the pricing will not help HP's revenue numbers, if it was deployed all the way down the customer lineup. A load that normally requires a $3.3 million system of 400 servers, with 10 storage racks and 1,600 networking and power cables using 91 kilowatts of power, could be done in the new system for $1.2 million: using one-half a storage rack, 41 cables and 9.9 kilowatts. Those are enticing number for cloud compute suppliers like the emerging manufacturing alternative Force.com. HP said these kinds of customers will be looking at system-on-chip solutions -- which could drive down the costs of cloud computing for the masses who might be migrating.
HP is tying Redstone into its HP Converged Infrastructure to help cloud providers share technology resources across thousands of servers. The technology targets the future of "low-energy computing for emerging web, cloud and massive scale environments."
In yesterday's podcast about the "post-HP" era, we pointed out how much spadework and mucking out the company must do to become relevant once more as an enterprise alternative to the 3000. Reducing power needs by 90 percent -- for anyone on the IT power food-chain -- qualifies as a leap out of the mud of "do we keep selling PCs, or not?"
If IT planning involves the selection of a vendor for your migration -- or simply a check to see if your application can operate on HP's environments -- this ground-breaker might deserve some time for a closer look, just to have something to quiz your cloud supplier about. One place to start is the "media kit" for its Low Energy Server Technology. What drives these innovations is a long way upstream from the architecture of a local server running MPE/iX applications. But this is the future where HP is investing in its hardware, "warehouse computing." An IEEE paper explains that power is one of the most critical concerns in this new concept.
One of the biggest trends in the server market has been the emergence of the large-scale data center, driven by Internet-sector growth. Indeed, recent market research identifies the Internet sector as the fastest-growing segment of the overall server market, growing by 40 to 65 percent every year, and accounting for more than 65 percent of low-end-server rev- enue growth in 2006. Furthermore, several recent news articles and keynote speeches have highlighted this area’s importance.
One of the most interesting aspects of this growth is the unique set of challenges it presents for server design. Internet-sector infrastructures have millions of users running on hundreds of thousands of servers, making the ability to scale-out server configurations a key design requirement. Experts have compared these environments to a large warehouse-style computer, with distributed applications such as mail, search, and so on. For companies in this market, data-center infrastructure — including power and cooling — can be the largest capital and operating expense, motivating companies to focus on the sweet spot of commodity pricing and energy efficiency.
Another IEEE paper points out the problem that the ARM-based Moonshot must overcome: latency. Software has to be explicitly parallelized to run effectively across so many "wimpy" processors, which adds to development costs. HP's going to open up a Discovery Lab in Houston in January and invite developers to test its new technology.
But they'll need to be clever to keep the feet of the IT horsepower moving on lower power.
The IEEE paper cautions that the overhead of splitting things up across multiple processors can reduce performance; and there’s the likelihood of lower utilization of the processors, and thus a loss of efficiency.
Once a chip’s single-core performance lags by more than a factor to two or so behind the higher end of current-generation commodity processors, making a business case for switching to the wimpy system becomes increasingly difficult because application programmers will see it as a significant performance regression: their single-threaded request handlers are no longer fast enough to meet latency targets. So go forth and multiply your cores, but do it in moderation, or the sea of wimpy cores will stick to your programmers’ boots like clay.
