A server industry group led by Intel is laying the groundwork for more so-called micro servers: tiny, power-efficient machines designed to handle light workloads such as entry-level Web hosting.
Last week, the Server System Infrastructure (SSI) Forum, a special interest group serving the x86 server industry, released the Micro Module Server Specification 1.0. Micro servers based on the specification “are just around the corner,” said Kevin Huiskes, Intel marketing manager in the data center group.
The bloom is off the rose for blades. They're dense and provide good performance, but they also consume a lot of power.
Pete Sclafani, CIO at 6connect
Work on the specification was spearheaded by Intel, Taiwanese manufacturer Quanta Computer Inc. and x86 server motherboard maker Tyan.
In fact, micro servers are arguably already shipping, said Jed Scaramella, senior research analyst at IDC. IBM, for example, has the iDataPlex, Hewlett-Packard has the ProLiant SL and Dell has DCS, a.k.a. “Viking.” Because of their size, these vendors tend to work directly with Intel, Scaramella said; the availability of SSI’s micro server specification, meanwhile, should help accelerate time-to-market for tier-two vendors with limited R&D resources.
What’s a micro server?
In terms of density, a good example of what an upcoming micro server might look like is the Dell Viking, which packs 12 single-socket micro servers in to a 3U chassis, said Huiskes. Generally speaking, micro servers will ship in a fully populated chassis that provides a shared power supply and fan like a blade chassis. But unlike blades, to keep down costs, a micro server chassis will probably not include any integrated switching or management.
Done right, micro servers could succeed where traditional blade servers have not, said Pete Sclafani, CIO at 6connect, a data center consulting firm.
“The bloom is off the rose for blades,” Sclafani said. “They’re dense and provide good performance, but they also consume a lot of power.” Micro servers stand to do well if they can provide similar levels of performance, but improve on blades’ power consumption and heat generation.
The performance offered by micro servers will be determined by the processors vendors put in them, which the SSI Forum spec does not prescribe. “Single-socket Xeons based on the E3 series is probably the bulk of what we think is the sweet spot for microservers,” said Intel’s Huiskes, referring to the upcoming Intel E3-1200 based on Sandy Bridge. But he also expects to see micro servers based on Intel ATOM chips, and potentially ARM processors.
The SSI specification also leaves vendors open to stack their micro servers horizontally or vertically. “The primary focus of the spec is on the ’pin out’ and the form factor of the card,” said Huiskes. “That leaves a lot open for the OEM to innovate around that board.”
Micro market for micro servers
Above and beyond what form micro servers will take is the question of who will adopt them, and to what extent?
Thus far, demand for super-efficient, dense servers was met by niche vendors like Cirrascale, formerly Verari, said IDC’s Scaramella. Specifications like SSIs, as well as the presence of startups like SeaMicro, suggest that there is interest in dense, low-power servers, but the market is still too small for IDC to track it, he said.
Indeed, in Sclafani’s experience, very few data center operators are open to talking about alternative server platforms. “A lot of customers like to talk about low-power options, but they tend to have strong vendor preferences.”
If micro servers do take off, it will be to host extremely lightweight applications that service many users, Scaramella said, for example, the login screen for an online gaming platform. While the game application itself runs on a full-fledged server, a full server is overkill for the login process.
But micro servers’ power efficiencies and densities come at a cost. Micro servers tend to be “stripped of any bells and whistles,” Scaramella said, including any hardware-level redundancy. Deployed by the thousands, micro server operators shouldn’t care “if they die on the vine. … Someone just comes in once and month and swaps out broken parts.”
That’s where virtualization and cloud technologies step in, said 6connect’s Sclafani. “There are two schools of thought about redundancy. One school wants redundant everything,” he said. The other school of thought is looking to virtualization technologies like live migration to “help reduce our redundancy dependency on hardware.”