Skip the navigation

Q&A: GM's CTO weighs in on grid

Ian Foster talks with Tony Scott about the automaker's grid work

By Ian Foster
January 28, 2005 12:00 PM ET

Computerworld - As I've noted in previous columns, one of the largest hurdles for the adoption of commercial grid computing is the enterprise IT community's perception that grid is still "just a science experiment." While we're seeing some substantial grid projects in place in compute-intensive vertical industries such as financial, pharmaceutical and energy, the typical enterprise IT professional still believes that grid hasn't yet proved its mettle for more mainstream enterprise environments.
In order to get a further sense of an IT buyer's thoughts about grid in the enterprise, I went to the top. I recently caught up with
Tony Scott, chief technology officer at one of the largest IT consumers in the world, General Motors Corp. One of the top companies on the Fortune 500 list, GM spends approximately $3 billion annually on IT. Here's what this global IT leader had to say about grid in the enterprise.
What needs to happen before grid is ready for use in more mainstream enterprise settings? Look at the core applications that are used to run most large companies. I'm talking about the bread-and-butter stuff that we use on a daily basis -- core financial systems, core HR systems, CAD systems and applications we use in other product development. Today, these are not written to the grid environment, which causes grid to have less applicability than it potentially could. We haven't seen the big application developers write to that environment. There are efforts that are laying some of the groundwork for that, but in terms of mainstream IT, we're not quite there yet.
Also, if we look at the management tools from the traditional systems vendors, we don't yet have the controls you'd need to effectively manage a grid environment with respect to commercial applications. They're coming, but they aren't yet at the same level of maturity as for our other traditional environments.
There are also issues with application scale-up. Take network capacity as an example. The industry still hasn't figured out how to effectively make network technologies scale up and down. It's still a fixed-capacity game in most cases, although MPLS [multiprotocol label switching] is helping there. But there's still a ton of work that needs to be done.
We're hearing about IT progressing from symmetric multiprocessing and mainframes, towards more scaled-out environments of commodity hardware. Is that trend reflected in GM's IT environment? For the Web environments, that is absolutely the case. Most large enterprises have the hardware applications within the Web environment that support that scale-out model nicely. But if you take more classical applications -- such as financial applications, human resource applications or other core IT applications -- again, the trend is less true in those areas. The whole architecture and model for how you run those was developed 10 or 15 years ago and doesn't embrace the scale-out concept from a model perspective.
Does GM have any pilot grid projects? Here at GM, we've used grid and its close cousins for several things, including crash simulation. In these types of specialized areas, it's been very applicable and valuable to us.
We're also doing some pilots right now of grid video delivery to the desktop. On the desktop, there is a shared directory. Once someone in the local network downloads a video, the next person in the local network who wants the same video can retrieve it locally, rather than clogging the WAN. It's simply a caching technique to get local stuff, but it's an example of grid and peer-to-peer that I think starts to address some real business issues.
In the old days, before peer-to-peer computing was as mature as it is now, we would never have been able to afford to do video to the desktop, because it would have killed our network. We've started to think about applying the same techniques to other areas. If it works for video, we could also distribute large CAD files that way. And we could distribute Microsoft updates that way -- it offers a way to deal with the real bandwidth hogs.
You've been pretty outspoken on identity management themes in the past. Do you think grid will further amplify the importance of identity management? It's evident to me that identity management is of increasing importance as we move forward in time. The days of relying on physical access to guarantee the security of the IT environment are ancient history. The ability of the typical firewall to keep everybody out except those physically on the network is also ancient history. We're quickly moving to a world where on an application-by-application basis, a subfunction-by-subfunction basis within applications, and in the entire IT infrastructure for that matter, doing business will require having a better awareness of who people are, what resources they should have access to, during what time frames they should have access to it and for what purposes.
The days of the sort of big bulk moves, where you say, "Everybody who is green has access, and everyone who is purple doesn't have access," are gone. The granularity of business today doesn't permit that. Our system of checks and balances is much more fine-grained than it ever used to be.
Now in the grid space, where you're allowing computers to participate in a community and share idle resources, your ability to do so will only be as great as your ability to protect those resources you need to protect. And as soon as you can't reclaim the resources when you need to use them, that will end your enthusiasm for participating in the grid.



Our Commenting Policies