Is Cloud Computing Overhyped?

Is Cloud Computing Overhyped?
Publication:www.cxotoday.com
April18th 2009

Barely has a section of India Inc. given nod to the cloud computing technology, a bigger cloud has emerged which can put efforts of cloud computing vendors in jeopardy.

We are talking about a recent McKinsey report that calls the entire concept a crazed overhype.' According to McKinsey analysts, a lot of companies are boasting about how they're cloud computing players, but there's still no real definition of what cloud computing is. They say that the buzz is disconnected from reality, and that the promised cost-savings are just not there.

But McKinsey is not all that hard on the technology. It says that while cloud computing may be more expensive for bigger businesses than the traditional data centers; it's good news for startups, and small and medium businesses.

But Yahoo!, the second largest search optimization company, may disagree. A major user of cloud computing technology, Yahoo! operates multiple cloud computing clusters that include clusters/grids of 1,000s of commodity machines. And also, 10,000s of machines running Hadoop (largest cluster has 1,600 nodes), TBs, RAM, PBs disk.

Hadoop is a software layer responsible for distributing application data across the machines, parallelizing and managing application execution across the machines, and detecting and recovering from machine failures.

Rajeev Rastogi, VP & head of Yahoo! Labs, Bangalore, told CXOtoday, "Many search/advertising applications are computationally intensive and have a need to process petabytes of data. This is only possible with large machine clusters with thousands of computers. Our Web search crawlers index several billion pages (100 billion+) distributed across the Web, and this is typically carried out in massive data centers containing thousands of machines."

Hadoop has been used to construct WebMap (graph of Web pages and directed links between them), compute Page Rank over the Web graph, and build the database of Search Assist (suggestions provided when using Y! search) from three years of log data.

Clouds are also useful for constructing machine learning models for predicting click probability from Web click logs in order to determine which ads (with the highest click probability) to place on a Web page, or in response to a search query, taking into account user behavioral, demographic, and geographic data.

Web pages are frequently clustered/classified to improve search relevance/quality/ranking or improve matching of ads with pages. Scaling to such large volumes of pages requires cloud computing support, said Rastogi.

Services firms too are testing the cloud computing technology. Gurdeep Singh, Head IT, Hanmer|MS&L, a public relations firm, has adopted the technology to host hanmermsl.com mails on Gmail cloud. "It saves money, server maintenance, and gives reliability of service," says Singh.

Our own CXOtoday survey, conducted as a part of the year-ender series, showed that a lot of CXOs are open to testing the technology.