Grid computing uses spare CPU power

The goal of grid computing, which gets its name from its gridlike architecture, is to link surplus computing power and other spare IT resources with clients who have periodic needs beyond the capacity of their machines.

There are four factors behind the growing interest in grid computing: the evolving of key standards such as TCP/IP and Ethernet in networking; the ever-increasing bandwidth on networks reaching into the gigabit range; the increasing availability of idle megaflops on networked PCs, workstations and servers; and the emergence of Web services as a logical and open subdivider of software computing tasks.

Grid computing software divides a task into subtasks, finds spare processors and other critical resources on the network, distributes the subtasks, monitors their progress and restarts any subtasks that fail. Finally, grid computing engines aggregate the results of the subtasks so the job or task can be completed.

One type of grid computing arrangement is a local cluster, which typically uses one main grid server on a single very-high-speed network. The grid machine handles one major task, and a small set of users are allowed to manage that task. A broader group of users are allowed to inspect and review intermediate and final results.

The next step up is the grid campus. Typically it involves many grid servers and many tasks. However, all the processing is done behind a firewall and network speeds are still fairly fast and within a known range.

Yet another approach is a global grid, which opens usage to machines anywhere on the Web and/or other private networks. It requires considerable effort to discover available resources and schedule tasks on these machines because they can differ so much in response times because of Web and network latencies.

Also, because traffic goes over the Web, security becomes a major administrative task, as does scheduling and monitoring dozens if not hundreds of grid programs and their legions of users with diverse roles and access privileges.

---

How it works
Subscribe to the Tech Update newsletter
Here is a weekly newsletter to help you stay abreast of new networking standards and technologies by providing down-to-earth explanations of how they work.

---

In terms of the impact on networks, grid engine software has reached a very high level of scheduling, dispatching and monitoring sophistication, so if a grid needs to cede a machine or clusters of machines back to end users, it can do so transparently and quickly.

The impact on network capacity depends on the nature of the application, but bandwidth monitors can be put into place to guarantee levels of standby network capacity.

The impact on applications is more difficult to characterize. For applications that need to be run in their entirety on a spare processing node, grid computing will work. But applications that need to be distributed over several processing nodes require some setup or reprogramming.

First, there is a class of applications known as embarrassingly parallel. These applications are easily broken into discrete computing subtasks that have little or no dependence on each other. These applications require some scripting setup, but many grid vendors provide either templates or simple routines that help to automate setting up the parallel workflow scripts required to run them on the grid.

Next there is a set of applications that have complex interdependencies known as lattice, tree or other iterative structures. These require significant interprocess communication and messaging, and are sometimes referred to as being codependently parallel. Preparing them for the grid is more challenging. For some of them, vendors already have produced parallel versions. And new standards are quickly emerging that will simplify producing parallel capable software that runs on grid engines.

This leaves a set of applications that have to be reprogrammed for parallel operations. Work is being done to automate this process.

The Global Grid Forum is working on a standard, the Distributed Resource Management Application API, that could improve grid applications development and interoperability by providing a standard API interfacing to an array of grid engines and operating system platforms.

Surveyer is a consultant and writer in Toronto. He can be reached at jbsurv@sympatico.ca.

Related Links

NWFusion offers more than 40 FREE technology-specific email newsletters in key network technology areas such as NSM, VPNs, Convergence, Security and more. Click here to sign up!		New Event - WANs: Optimizing Your Network Now. Hear from the experts about the innovations that are already starting to shake up the WAN world. Free Network World Technology Tour and Expo in Dallas, San Francisco, Washington DC, and New York. Attend FREE		Your FREE Network World subscription will also include breaking news and information on wireless, storage, infrastructure, carriers and SPs, enterprise applications, videoconferencing, plus product reviews, technology insiders, management surveys and technology updates - GET IT NOW.