Intel on Tuesday said it will offer a range of Sandy Bridge-based micro server platforms and expand its current Xeon family products as the company predicts that the micro server category will grow over the coming years to occupy up to 10 percent of the server market.
Intel's upcoming processor offerings for the micro server segment include its Xeon E3-1260L processor running at a TDP of 45W and the Xeon E3-1220L running at a TDP of 20W, both of which are in production; an as-of-yet unnamed Sandy Bridge-based micro server processor running at 15W TDP due in the second half of this year; and, finally, an as-of-yet unnamed Atom-based micro server processor running at a TDP under 10W due next year.
Boyd Davis, general manager of Intel's Data Centre Group, at a press briefing in San Francisco, said there is going to offer a range of products for highly-dense, low-power architectures aimed at the emerging cloud-based Web service industry.
"We're pretty excited about the micro server category for specific workloads," he said. "There is a theoretical opportunity that this class represents, but it probably won't reach that because there are other technologies that meet that need, such as virtualisation. But the challenge with virtualisation is virtualizing at scale, which means smaller, individual nodes have a role to play."
Davis said Intel chose to introduce high-density category servers distinct from rack-mounted servers as the blade category evidently did not grow as quickly as the rack segment, and as micro severs enable environments where multiple modes share power cooling, reducing power consumption at the individual node level.
Boyd said Intel expects for the majority of the market to be occupied by Xeon as the micro server category grows, it will grow first and fastest on Xeon," he said. "But we'll have a solution for customers who see their workload running on Atom servers."
Davis said Cloud service providers are where the sweet spot of this market is.
"Almost all the volume for micro servers come from cloud service providers in some form," he said. "A lot of applications are optimised to have a pretty beefy node. The key is to have technologies that address the highest possible market need within that ten percent, but is versatile enough to offer low-power and high performance for a broad range of workloads."
For that reason, Davis says, it's critical for any platform to have the ability to virtualise. He said that if Intel can offer instruction set compatibility for older generation x86 products and ramp up performance power on its upcoming chips, it makes sense for customers currently leveraging x86 technology to stay there.
Next: Systems Running Intle Micro Server Products
Davis said that establishing an industry ecosystem around its platform is critical to Intel's efforts in the micro server space. He cited SeaMicro's release of an Intel Atom-based server earlier this month, and said that companies including Dell and Tyan are driving Intel's momentum around micro servers with systems based on Intel's Xeon line.
"We're a building block supplier," Davis said. "We can provide a certain amount of building blocks but it's critical for companies looking to serve the cloud market in a big way to have a very consistent view with Intel on how this market is going to emerge to meet in order to meet the needs of our customers."
In addition, Intel said its Micro Server Evaluation Lab will debut in the second quarter, offering software developers to determine the best Intel technology options for their own usage models and applications. Davis said it may not be obvious how various applications will behave when developers try to run them on various architectures. Davis said the Lab will be accessible online in the next 30 to 45 days.
"We're excited about the micro server business," he concluded. "It's not going to take over the universe but this is a critical segment and we're going to win it."
Gio Coglitore, director of Facebook Labs, spoke at the event as an Intel customer, eager to leverage opportunities in the micro server segment. "We believe in the direction Intel has taken with its micro server architecture," he said.
"Facebook believes that there's no single solution that fits all, whether it’s a Sandy Bridge-based or a small form factor micro server. What excites us is we see several different directions coming together, from the downgrade SeaMicro has done to the standard approach Dell is taking to the market."
Coglitore said the micro server market is beginning to be taken very seriously, as there are certain workloads that benefit from the focused approach that micro servers bring to the IT environment. He pointed out that micro server deployment cannot always be virtualised, and that Facebook prefers a "realised environment" as it seeks to scale its servers out further.
"The micro server allows us to target that particular workload and scale like we haven't been able to before," he said.
Coglitore said Facebook's front-end workloads cannot be virtualized, and must therefore rely on a realised environment, with new products deployed quickly -- which Intel's tight "tick-tock" roadmap strategy offers.
"Facebook is on a very rapid compute cycle of two or three years," he said. "So it isn't necessarily that you can't virtualise but a realised environment is the right approach for us, across our whole infrastructure. So we really like Intel's tick-tock approach."
Coglitore said it can be very expensive to keep the number of cores on a system to a reasonable amount as one scales out, but the key is to bring flexibility to the market and allow customers to find the perfect solution. For this reason, he said, Facebook has tested micro servers in production, and is looking to deploy them in masse sometime later this year, or in 2012.