How Cloud Computing Is Transforming The Mega Datacenters

Lately, we read a lot about the NSA (National Security Agency) U.S. be accessing the major service providers servers on the Internet. Too much water will still roll until the case is properly clarified and exit the field of speculation, but the theme has opened my eyes to a phenomenon that these companies have set: the mega datacenters.

These huge mega data centers are quite different from traditional datacenters we know, even the most major banks and industries. A mega datacenter has at least 15,000 servers, and cost at least another $200 million to build.

They have a common characteristic: Automation to the extreme. And can scale massively and employ few people to be managed. This means you can add a few thousand servers (or tens of thousands) with an increase of almost negligible cost.

Its operating model will, in my opinion, spread by other datacenters and increase attention to the cloud computing model. The mega datacenters will not end with traditional datacenters – at least in the foreseeable hoorizonte because they were not built to workloads from legacy systems. Were designed for typical Internet workloads such as Facebook, Google and others. Recalling that cloud is essentially a limit on automation with software managing the whole operation, detecting and recovering from faults automatically managing the inclusion or exclusion of new servers, automatically installing new versions of operating systems and so on.

A new dedicated server is put into operation in a few hours. Its growth rate is fantastic. Each round of delivery may reach a thousand servers. For example, Facebook had approximately 30,000 servers in 2009 and in late 2012 was already between 150 thousand. Google, meanwhile, had about 250,000 servers in 2000, 350 thousand in 2005 and 2010 has reached to 900,000. Today must be over one million servers! How many companies in the traditional world, acquire lots of thousand servers at a time?

Automation (cloud model) is essential for the operation of these mega datacenters. Generally, in traditional datacenters there is a relationship of a sysadmin for every 100 to 200 servers. On Facebook, a sysadmin compared to 20,000 servers. This means that a sysadmin on Facebook does the job of 100-200 professionals from traditional datacenters. As the infrastructure management is automated, we see that the main actor in this process is the software. Pinterest, for example, when it had 17 million unique visitors per month, held just a sysadmin to manage cloud. Staff cost us mega datcenters nor is among the Top 5 items of cost, unlike traditional datacenters.

The services provided by these mega data centers and its economy of scale allow the creation of new businesses unviable in the traditional model of upfront investment. An example is Netflix, which manages more than 36 million users video streaming in the public cloud solutions. In the first quarter this year went over three million new users. Other typical internet companies world could only grow the way they grew up using the economy of scale provided by mega datacenters, like Instagram, Zynga, FourSquare and Pinterest.

Pinterest is an interesting case. Spent twenty terabytes of data stored to 350 in just seven months, using a cloud on a mega datacenter. In the traditional model of purchasing physical servers would be absolutely impractical to achieve this expansion in a timely manner.

Investments in its creation are outside the curve. Some estimates suggest that the costs of building some of the datacenters are fabulous. For example, Facebook mega datacenters cost $ 210 million in the U.S. state of Oregon, reaching U.S. $ 450 and U.S. $ 750 million in the other two new, state of North Caroline. The Apple in Oregon cost $ 250 million. Google is not far behind: $ 300 million in the mega data center in Taiwan and $ 1.9 billion in the state of New York. Microsoft, in turn spent $ 499 million on its mega data center in Virginia and the NSA, National Security Agency of the U.S. government, is building a mega datacenter $ 2 billion in the state of Utah.

Well, way these mega datacenters are created? Basically to meet public cloud offerings and Internet services direct to consumers (B2C). Investments in capex of mega cloud providers is immense. It is estimated, for example, that the capex Google reach the level of $ 1 billion per quarter.

Interesting that the mega datacenters are creating a new industry in the IT industry. In general, they do not buy traditional suppliers servers, but use other sources, based on own and assembly performed in Chinese companies such as Quanta Computer designs. Quanta, for example, provides 80% of the Facebook servers and is a leading supplier of Google. This model leads to a paradigm shift. For the mega datacenters, it is much cheaper to exchange server to fix it and thus the time to buy machines with higher index MTBF (Mean Time Between Failures), prefer cheap and disposable machines. If there is a problem, the automation software simply puts the server crash out of the air and reconnects service in another, without human intervention. On the other hand, this is one of the reasons that this current model cannot be applied in traditional datacenters: the software that runs on them, such as public Google services were designed to operate in seamless integration with automation software. Very different from a corporate environment, where a SAP was not designed to work seamlessly with any vendor servers.

The mega datacenters adopt magic formula of cloud: Virtualisation + Standardization + Automation. The standardization facilitates this replacement, since each box is equal to another. The same version of the same operating system version and the public software is operating on all machines.

Recently, Facebook opened its black box. Adopting described five types of servers, one for each specific activity. Basically, it divides its servers into five types: those that meet Web, database, Hadoop, and Haystack Feed load. Each of these services demands different settings. For example, a server that handles photos and videos need more memory than computing power.

The mega datacenters represent the phenomena of consumerization to the traditional data center model. Buy machines by the thousands and are redefining the industry of Intel based servers. An Intel study shows that in 2008, 75% of processors for servers were manufactured by IBM, HP and Dell. Today, these 75% are ditributed from eight manufacturers and interestingly Google is already the fifth “manufacturer” of servers on the market. Really, we are experiencing a disruption and arguably the cloud computing model that has much to do with these changes in the industry.

Be Sociable, Share!