Intel Sees Itself Between the Clouds
Page 1 of 1
SAN FRANCISCO Like any new market sector taking shape, cloud computing has a large number of players jockeying for position.
IBM, Microsoft, VMware, Salesforce.com (NYSE: CRM) and Amazon (NASDAQ: AMZN), as well as smaller players, are pushing for control in the emerging sector; Intel (NASDAQ: INTC) plans to be there with its own products and technologies but taking no sides.
"Our role is we get to be a partner to all of the big companies, as opposed to competing with them," said Jason Waxman, general manager of Intel's High Density Computing group. "None of them view us as a competitor, they view us as an enabler."
The company defines cloud computing as consisting of three elements: the architecture, the services, and public vs. private clouds. The architecture is the most important part, and that is seen as stateless and not tied to a single server.
"If the Intel Exchange server goes down, I don't get e-mail until that particular server is back up and running," he explained. "With a cloud service like Yahoo or Google, because it's stateless, I get my e-mail on a different server every time."
So applications are no longer tied to a specific piece of hardware. There are no "Exchange servers" or "application servers," as the services move around on hardware. As needs scale from increased user demand, the service can expand out across multiple machines to bring more compute power to bear.
All of this is handled by the second part, services, which deliver cloud services to users.
The third part is private vs. public clouds, with services like Google and Facebook falling into the public domain and internal business use falling under private.
Surprisingly, Intel thinks the SMB market is better suited to begin adopting and deploying cloud features while the enterprise needs to go slower. Small and mid-sized businesses, there are a few attractive elements to cloud computing, namely the chance to avoid deploying a widespread infrastructure. There, Amazon S3 or Salesforce.com might be a better choice.
On the enterprise side, Intel's advice is more "approach with caution." Companies can begin to evaluate the cloud and begin preparing, but a wholesale move to the cloud is not recommended, because larger enterprises tend to be locked in to technologies. How often, for instance, do enterprises jump from one database vendor to another? The same applies with the cloud, and the cloud as a market has not shaken out yet.
What Intel does best, engineering
On the technology aspect, which is what Intel excels at, the company is doing its best to help with datacenter issues of power and cooling. That includes using high efficiency voltage regulators on the motherboard to improve power management and changing the layout of the motherboard for optimal air flow.
Intel also redesigned its motherboards to remove all unneeded components, like sound and video chips and expansion slots and ports. That keeps the power draw down and makes air flow more easily. Using a Core i7-based server, Waxman said it's possible to get a server down to just 85 watts of total power when idle and 250 watts under load.
Solving the problem of I/O, virtualization's biggest bottleneck, is coming through a combination of hardware and software. The cost of 10 gigabit Ethernet is coming down, he noted, and IBM has its own technologies, IOVirt and VMDQ. VMDQ, the virtual machine device queuing technology, manages and prioritizes data so the data from multiple virtual machines don't all try to go through the same I/O port at the same time.
Another Intel technology is Node Manager, which intelligently manages power in a rack server. Working with the Chinese search engine Baidu, Intel was able to save about 40 watts per server.
Finally, the company plans to promote regular air cooling, following the success of its New Mexico experiment, where it found that just using ambient air temperature was just as efficient at cooling a datacenter as using an air conditioner to drop temperature to 50 degrees.
With all of these tricks, Intel noted the savings. In a 50,000 server datacenter, a 10 percent improvement in motherboard voltage regulators translated to a $6 million per year savings. Node manager knocked off 30 watts per server, saving another $8 million per year. Thanks to a 10 percent improvement in software parallelization, $20 million was saved, and finally, a 10 percent improvement in power usage effectiveness (PUE) resulted in a $1 million savings.