Grid Promises, Realities

What is vital now is the constructive skepticism of potential customers.

This weeks OracleWorld Conference in San Francisco will swell the ranks of enterprise applications for the grid computing model. With the promised debut of Oracles Version 10G database and application server, enterprise buyers will be able to ask their Oracle representatives—as well as their IBM contacts, who already carry impressive grid portfolios—to move beyond praise for the potential of dynamically allocated resources and to start talking seriously about return on grid investments.

Its a trivial analogy to say, as proponents affirm, that a computing grid will discover and allocate the perishable resource of computing power—in the same manner, essentially, that the electric power grid allocates the equally ephemeral worth of watts. Less often mentioned are a computing grids added burdens of protecting interactions against intrusion, interception and disruption. Volts and amperes dont need encryption to protect against sniffing or spoofing; neither do they depend on time- and sequence-sensitive routing among processes.

If the computing grid is going to make effective use of specific processing capabilities or particular data collections and availabilities, there is a need for sophisticated protocols to make each nodes needs and capabilities known to the others.

Without large-scale solutions to these problems, a computing grid is even more certainly destined for failure than the nations demonstrably fragile power network. Buyers should look for these answers as a precondition.

Once buyers have made clear their demand for robust enterprise solutions and their ability to recognize a research project in products clothing, its vital that the resulting products be implemented according to nonproprietary standards. The key argument for grid computing is one of buying computer power in bulk and allocating it to individual tasks with low overhead and at minimal marginal cost. That argument defeats itself if it comes to market in packages that tie buyers to vendor-specific protocols.

We agree with Oracle Vice President John Magee when he says that "when you look at these commodity clusters, a big, fat expensive operating system sticks out like a sore thumb"; we note with approval that IBMs grid efforts also treat nonproprietary Linux as an important component. We welcome the fervent promises we hear from vendors for efforts such as the Globus Project and the Global Grid Forum; were pleased to hear Hewlett-Packards chief scientist, Greg Astfalk, call for "an open-source, community effort."

What is vital now is the constructive skepticism of prospective customers in the face of vendors likely efforts to sugar the pill of platform lock-in with promises of better performance and quicker time to market. Were prepared to believe grids can be made to work; we invite IT developers to convince us that grids can become a competitive market and an economically compelling approach.

Send your response to