Using a micro-server to handle massive amounts of data-do you dare to try?

Recently, eharmony, which provides online chat services, has purchased a batch of Seamicro servers with built-in Intel Atom processors to run distributed computing services based on the Hadoop architecture.

If it was a few years ago, eharmony on the server with the Atom processor behavior is tantamount to suicide, after all, atom performance is limited, and the industry does not have a complete microprocessor solution. But now the servers developed by Seamicro and Calxeda, which mainly use the ARM processor's micro-server, have been able to significantly reduce construction costs and save energy.

However, the industry generally believes that how to meet the needs of applications and large customers is the first problem that these micro-servers need to address before they become popular. For large data processing, a processor like atom needs to solve all two problems-

Calxeda: Using arm because there is no historical burden

Although Calxeda's ARM architecture micro-servers are very different from the processor architecture of the Seamicro Company's X86 architecture, Evans's first two advantages are also applicable to Seamicro X86 architecture products.

"For massive amounts of data, there are three key factors that the ARM architecture server will be good for," says Barry Evans, CEO of Calxeda. First of all, the processing of the massive data has the outward extension characteristic, needs many high efficiency processor to carry on the processing. Second, the massive data processing as a new market, it has to face a few historical problems left. Finally, the current software for this area supports the ARM architecture very well. ”

Hadoop on Microservers

As a distributed computing structure, Hadoop is a new field for many companies and organizations, and most of them do not have the hardware specifically to deal with this application. Distributed computing requires a large number of scale out extension structures to enhance the parallel processing capability of the whole system. For each node of distributed computing, Hadoop does not require its performance to be comparable to a high power, high-performance, dedicated processor, and on the contrary, its performance requirements are very low. However, the micro-server can meet the requirements of Hadoop calculation by virtue of a lot of material core and relatively small volume.

Atom node minimum power consumption of only 5 watts

Seamicro's micro-servers can provide 512 Intel Atom cores in a 10U rack and generate 1.28tb/s computing power, compared to traditional servers, where energy consumption is reduced by 75%. The Calxeda Company's micro-server products can provide 120 4 core ARM Cortex A9 architecture processors in a 2U volume, while each processor node consumes only 5 watts. "Intel and AMD are based on the x86 architecture of the four core processor power consumption between 80 and 130 watts and the energy-saving version of the x86 chip also has 30 watts of power," said Stacey Higginbotham, a prominent journalist, when talking about the Calxeda plan. For those who insist on using traditional server chips, consider Dell's Cloudscale micro-server, which uses an Intel Xeon processor with a power dissipation of 30 watts.

Both Evans's theory and eharmony's example show that micro-servers are appropriate for large data calculations such as Hadoop. Before using the micro-servers, eharmony companies used cloud services to meet their needs, but they found that this approach became very expensive when they were doing a lot of parallel computing and exchanging large amounts of data with the cloud. As a company that has not previously invested in any server hardware, eharmony can easily switch to a micro-server architecture. Rich Miller, of Data Center knowledge, also said that "the micro-server can reduce the CTO's operating costs by tens of thousands of dollars a month, reducing the cost by 74%." ”

With the success of the eharmony case, we may soon see companies like Seamicro and Calxeda reach OEM agreements with Cloudera, the largest distributed computing software provider at the moment. Of course, with the micro servers catching up in this area, we are also likely to see providers of software and hardware integration solutions.

The future of the micro server

As you can see, the advocates of micro-servers are so optimistic, largely influenced by the booming Internet and cloud computing markets in recent years. Technology executives like Facebook say the company is not ready to embrace the trend of server virtualization, but rather to use it as a cheap growth and fast failover approach. Currently, Facebook has tested the micro-servers and is intentionally using them for large-scale data centers.

Data show that Facebook in June 2010, the number of servers exceeded 60,000, and April 2008 10,000 units, even to the end of 2009 is more than 30,000 units. And as the Oregon Data center opens this year, Facebook's servers will be close to 100,000.

Facebook may be a special example, but it does represent a trend for fast-growing emerging internet companies. In the future, when the emerging internet business and cloud computing needs are completely awakened, it may be the real spring of the micro-server.

The latest news is that in the next few days, including MIT renowned Professor Anant Agarwal, is also the Tilera company's founder and CTO. and Partha Ranganathan of HP Labs will begin to delve into the next generation of distributed architectures, and the purpose of this meeting is to explore the concept of compressing a large number of low-power processors into small processors for applications with massive data outside (primarily network applications).

(Responsible editor: admin)