Internet Scale Computing

Short excerpt from a paper to be published soon. Feedback welcome.

The only constant in the world of technology is change. Service Delivery in both the enterprise and consumer space is undergoing significant change and the line between them is becoming blurred, enabling each of these domains to take advantage of the technologies of the other. Traditional enterprise solutions such as line of business applications (LOBs) and service registries are starting to incorporate concepts associated with "Web 2.0" such as AJAX, wikis, community feedback and other popular trends. Likewise, consumer services are looking to web services standards as a way to enable security, trust, non-repudiation and other requirements typically associated with enterprise solutions.

Extreme scalability in consumer services is critical since the targeted user base is far broader and more diverse than a single enterprise. Service interruptions in consumer services can be detrimental to the financial status of both service providers and service consumers. In late 1999 eBay was unavailable for a little over twenty-two hours and as a result suffered significant financial losses. Today we are seeing similar scalability challenges facing consumer-oriented service providers such as Skype, Amazon, MySpace and others.

The challenge of extreme scalability is further compounded by inefficiency and poor utilization of resources within the datacenter. Server farms are designed and built for maximum load conditions and dedicated to specific solutions, effectively locking horizontally-oriented resources into vertically-oriented “solution silos”. Average datacenter utilization rates typically range from 15-20%. These incredibly low utilization rates detrimentally impact infrastructure and solution ROI levels.

The emergence of broadly available consumer-oriented services coupled with requirements for extreme scalability and higher utilization rates leads us to a new computing paradigm we call Internet Scale Computing (ISC) .

Internet map created by the Opte Project. Click on the graphic for more info.

Update: Thanks Joe!