SeaMicro is coming out of three years of stealth mode with a new server architecture that offers significant power and space savings through its use of Intel’s Atom processors.
SeaMicro on June 14 unveiled its SM10000, which can scale to 512 Intel Atom chips, which Intel initially designed for use in netbooks and now is aiming at a host of mobile and embedded devices.
However, SeaMicro officials saw that the highly power-efficient chips could be used in servers targeting the burgeoning cloud computing and Internet workloads.
There is a shift in computing away from large data center workloads to environments where there are significantly more-but smaller-tasks, according to SeaMicro CEO Andrew Feldman.
Other systems makers-including IBM, Hewlett-Packard and Dell-are still making servers designed to handle those larger workloads, and use up too much power and space running the smaller ones, such as checking e-mail, social networking and viewing Websites, Feldman said in an interview with eWEEK.
The Atom chips are much better designed to handle such workloads and in the process save businesses money in power and space costs, he said.
“On the Internet, the problems that need to be solved are much smaller problems,” Feldman said. “Atom is really good at these ordinary problems, but doesn’t do as well with the harder problems.”
Today’s Internet-based traffic also tends to be “bursty” – it rises and drops quickly, he said. The result is underutilization of traditional servers. However, the design of the SM10000 makes it much easier to scale, Feldman said.
SeaMicro is using the Atom platform and a patented I/O virtualization technology to shrink the motherboard by eliminating about 90 percent of the components from it. The capability not only drives down the size of the motherboard but also the power consumption.
Feldman said the motherboard is shrunk from the size of a pizza box to that of a credit card.
An interconnect technology links 512 of these motherboards to create a single clustered systems that offers 1.28 terabits-per-second of throughput. The architecture also can support a variety of processor instructions sets, such as ARM, x86 and RISC, and protocols that include Fibre Channel and Ethernet.
SeaMicro offers server management through its DCAT (Dynamic Compute Allocation Technology), which through dynamic management and load balancing ensures that the myriad chips in the system are run at their most energy efficient.
SeaMicro can put all of its compute, storage, interconnect and server management technologies into a 10U (17.5-inch) server.
The result is a server that can offer high performance in a form factor that is a quarter of the space and power of traditional servers, Feldman said.
SeaMicro is rolling out the SM10000 at a time when businesses are becoming increasingly concerned about power, cooling and space costs. Feldman said that for Internet companies, power is 30 percent of its IT operating expenses. He also pointed to an Environmental Protection Agency report that said data centers now account for almost 1.5 percent of the power consumed in the United States.
Chip makers like Intel and Advanced Micro Devices are looking to increase the performance of their processors while driving down the energy consumption, and OEMs are looking to integrate power saving capabilities their systems not only through the use of such chips, but also through other server technologies.
There also is a push to use other chip technologies, such as low-power processors from the likes of Via Technologies, and graphics technologies from AMD, Nvidia and others in servers.
Zeus Kerravala, an analyst with The Yankee Group, said SeaMicro’s offering is an interesting concept that, if proven, holds a lot of promise.
“You have the computing power of a regular server, but it’s just a bunch of smaller processors,” Kerravala said in an interview. “But at any moment, you can scale up or down to the exact number of processors needed. It makes so much sense, you kind of wonder why no one has thought about it before.”
Now the challenge for SeaMicro is to prove its architecture in the field. Kerravala said the offering initially will have particular appeal in environments such as Internet companies and HPC (high-performance computing) organizations, where there’s a high demand for compute power at lower power and space costs.
If it works well in those spaces, enterprises may come calling. And if that happens, Kerravala said, businesses can expect mainstream OEMs will start offering similar systems in their portfolios.