book.chapter Grasping cloud basics

The Internet was first described as a “cloud” where every computer would be connected from the very beginning. But more recently, the meaning of the cloud has changed. Today's cloud consists of massive data centers that are more standardized and automated than ever before. Rather than containing expensive supercomputers, these data centers use hundreds of thousands of inexpensive personal computers working in parallel. Highly sophisticated software balances the computing demands across these microcomputers instantly. The major players in cloud computing carefully conceal the size of their data centers to avoid setting off an arms race over who has the biggest cloud computer. But some facts were known in early 2010: Microsoft was building six data centers to power Azure, its cloud service. Two were in North America - Chicago and San Antonio. Two more were planned for Europe - Dublin, Ireland and one other location. The final two were in Asia - Hong Kong and Singapore. In each region, the data centers were linked to back each other up. Microsoft budgeted around $1 billion per region. The company used Intel's product where a shipping container of 2,000-2,500 pre-installed computers was delivered, plugged in, and activated immediately. Google was rumored to be constructing at least twelve massive data centers worldwide. Each would contain 500,000 to 600,000 servers. Google surpassed one million servers in 2009, although the exact number remains secret. Amazon was known to be building a $100 million cloud data center in Boardman, Oregon to access inexpensive hydroelectric power. The 117,000 square foot facility was surrounded by barbed wire fencing and tight security. Amazon was expected to build two more centers soon. In Virginia, Terralink built a $250 million networked government data center with reinforced walls, security cameras, and guarded gates. Clearly, much more was happening with these data supercenters than publicly revealed. Astute companies were making huge investments in high performance centers that would become integral to cloud computing. This level of spending signaled an important development was underway - a potential game changer. Mainframe computers were once expected to dominate computing. Later, expensive supercomputers were seen as the next big thing. But instead, numerous cloud data centers were being built from hundreds of thousands of inexpensive, mass-produced PC parts. Although a single desktop CPU was unimpressive, thousands clustered together in a data center represented a new, affordable, high performance computing resource. Clouds are essentially a supercharged form of cluster computing. Early providers of large public clusters include Amazon Web Services, Google App Engine, Microsoft Azure, Rackspace, Oracle (Sun), IBM, Yahoo!, eBay, and Facebook. More companies aren't actively involved because building a functioning cluster is technically challenging. Keeping numerous computers working in unison generates overhead to synchronize everything. Oversight systems track each node’s activities and prevent conflicts. Bigger clusters require more resources just for organization and management - estimated at 10-15% of total processing power. Various cloud software has been developed to handle these cluster management needs. The Internet was designed to survive nuclear attacks by routing around any failed segments. Similarly, cloud data centers have extensive redundancy. If individual servers fail, their data is allocated to others. By distributing risk across hundreds of thousands of servers managed by intelligent software, clouds can be highly fault tolerant. Cloud computing represents a fundamental shift in IT and a new computing distribution model. Clouds will come in three main types: Private clouds - Companies make their data centers more cloud-like for internal use. In 2009, Rackspace and Amazon announced dedicated servers for private clouds, physically segregated for security. More companies will likely build private clouds as the advantages become clearer. Hardware vendors now offer products for private clouds including Dell, Cisco, HP, IBM and Sun. Public clouds - Available to anyone online. Users pay only for resources used. In 2010, Linux servers cost 8-12 cents per hour on Amazon or 1.5 cents on Rackspace. Windows servers were 12-12.5 cents per hour. Prices will likely drop as competition intensifies, approaching the cost of electricity alone. Hybrid clouds - Private clouds using public clouds for overflow capacity during spikes in demand. Currently, many companies overinvest in IT to handle occasional peak loads, so their servers operate far below capacity most of the time. Hybrid clouds allow private servers to run at 90% utilization or more, offloading peaks to the public cloud. New developments are unfolding in all types of clouds. Paying only for computing power when needed is a key driver. Hybrid clouds synchronized with public resources could make corporate IT more cost effective than ever. Software is enabling seamless workload sharing between private and public clouds. And an avalanche of new cloud services and products is coming to market across industries. Excitement is buzzing about cloud computing's massive scalable power available on-demand. It represents a break from previous IT models. Developers are working on intelligent software and services to leverage the cloud’s potential. Those who harness its capabilities will pioneer a new digital culture and transform hopes and dreams into realities.