Archive for February, 2005

Grid Computing: Ready for the Enterprise?

February 16, 2005

While the champion New England Patriots were gathering outside with fans to celebrate their victory on the gridiron, the geeks gathered inside the Sheraton Boston from Feb 7-11th to tackle grids of a different sort at the 3rd annual GlobusWORLD Conference.

I attended the conference to check in on how much traction grid computing is developing in the mainstream, commercial computing world. It’s definitely coming—yet another example of work within the High Performance Technical Computing community laying the groundwork for mainstream computing advances–it’s just a matter of timing.

I attended talks on the Enterprise Track rather than any of the detailed talks on Globus technology or components. In my view, while the Globus community and the Globus Toolkit are key intellectual drivers in the grid space and are the focus of a thriving research community, it is unlikely that that Globus itself (at least in its current form — see the Univa website for a possible future for Globus in the enterprise) will be adopted by enterprises as an integral part of their key business processes. This view was bolstered by my experiences at GlobusWORLD.

What I’d like to do here is share some thoughts on what’s been happening in the grid computing world and why it matters for mainstream commercial enterprises. Along the way, I’ll give some capsule summaries of what I heard real enterprises like Sprint, SAP, and Hewitt doing with grid computing. And I’ll give you some pointers to a couple of software vendors with some very interesting grid computing technologies that are aimed at the enterprise.

First, what’s a grid and what is grid computing? At Sun, we talk about a hierarchy of grids:

  • Cluster Grids — A group of IT resources in a datacenter or single administrative domain within an organization.
  • Enterprise/Campus Grids — A group of IT resources, probably geographically distributed, certainly administered by separate groups within the organization, and all within the boundary of the organization’s firewall perimeter.
  • Global Grid — A group of IT resources, geographically distributed, owned by separate organizations, often communicating using the public Internet.

In all cases, it isn’t a grid unless the resources are being used collectively in some coordinated manner to solve problems in a distributed manner. Here are some typical examples of grid computing:

  • Rendering of movie special effects and animation
  • Genomic sequence matching and protein modelling
  • Derivative analysis and risk modelling for finance
  • Electronic Design Automation [example: Sun’s Compute Ranch]

At GlobusWORLD, several early-adopter enterprises offered glimpses into their use of grid technologies. For example, Hewitt discussed their integration of grid computing technologies (running on Linux-based clusters) with their existing mainframe infrastructure to deliver higher performance on two of their existing enterprise applications with high cost effectiveness. It was interesting to see technology approaches originally developed for use by rocket scientists (real ones) being used successfully for pension calculation as well as PDF and postscript generation. Hewitt worked with Data Synapse to create these solutions. I hadn’t heard of Data Synapse before the conference and intend to spend some time learning more about their technology — it looks quite interesting.

Sprint gave a talk on their technology explorations with grid computing. Though the speaker couldn’t give too many details of what they were doing for competitive reasons, he did mention that they had done an initial test with 30 low-end x86 nodes and then a subsequent transaction-based test using about 100 heterogeneous nodes that delivered about 6000 transactions/sec using 5KB packets and some arbitrary processing to simulate a transactional environment. The technology they chose for these tests is at least as interesting as that used by Hewitt: Hive Computing from (the unfortunately named, but soon to be re-launched) Tsunami Research. I found the philosophical underpinnings of their approach to be quite compelling. Worth checking out.

Presenters from Mazda, Wachovia, SAP and Bowne also shared their early experiences with grid computing and all expressed a belief in the approach though all had encountered the typical challenges of inserting new and evolving technologies into a corporate environment.

This entry is already too long so let me end with a quick note on the technical trajectory of grid computing. As implemented within the High Performance Computing community, early and current approaches to the various aspects of grid computing have been very traditionally oriented: lots of UNIX philosophy, FORTRAN codes, etc. The fact that commercial enterprises are starting to look seriously at grid computing approaches is owed to the work being done under the auspices of the Globus Consortium and other grid research groups to merge grid computing with web services and service oriented architectures (SOA). This is an important shift in direction that will prove pivotal in the widespread adoption of grid computing capabilities by mainstream commercial enterprises.