Grid computing is a noteworthy topic, particularly this week, with formation of the Enterprise Grid Alliance by Oracle, Hewlett-Packard, Sun Microsystems, and others.
Platform Computing has been providing software to enable development of grids and is watching the activities of the Enterprise Grid Alliance with interest. Recently, InfoWorld Editor at Large Paul Krill discussed grid computing with Platform Computing’s Ian Baird, vice president of marketing and chief business architect, to talk about the so-called politics of grid, standards, obstacles, and the possibility of markets emerging for compute cycles based on grids.
InfoWorld: What is your definition of grid computing?
Baird: For the purpose of the studies that we’ve been doing and for a broader definition around grid, because there’s still considerable debate about grid, I at this point am defining grid or working on a definitional document at the Global Grid Forum with a number of different industry vendors, trying to find a consensus definition. We refer to grid in a very broad sense [as] any sort of distributed or cluster computing that involves the virtual pooling of IT resources.
InfoWorld: What do you mean by the politics of grid and how this has changed recently?
Baird: Last year we did a study on the market and asked some 50-plus multibillion-dollar corporations, the CIOs and others, about the issues or barriers to grid computing. One of the biggest barriers that we discovered was that there is a group of what we call non-technical issues related to grid computing, and effectively that turned out to be organizational issues or what we described as “the politics” of grid. The issue there is really that people are unwilling to share their resources.
What companies built out, in many cases, is a very siloed world of compute resources, where one department or one group has their set of compute resources and therefore their particular jobs that support their particular business department and that’s it. And if you’re going to deploy a grid, then you want to start connecting these various departments and start harnessing the resources across multiple groups and tapping into the power that the aggregated virtualized consolidated resources bring.
In the case of politics, then, what happens is most people feel -- and this came out in our study last year and again this year -- is that the politics are a significant issue. Last year, 89 percent of the respondents to the study said that organizational politics were an issue for grid computing, that people are afraid that they will lose control or access to their resources. People felt that there are risks associated with enterprise deployment of compute resources, they felt the risk of loss of budget or dollars, loss of priority for their projects, etc. In those contexts then, there enters politics and people making reasons why they can’t share their resources, why they can’t aggregate them and virtualize them. And that’s what we call the politics of grid.
What we did then this year was a follow-on study to that study last year, and asked -- have the organizational politics around grid become more or less prohibitive this year versus last year? And the finding was in the study that 43 percent -- and this was of a new group of CIOs, some 100 CIOs that we surveyed -- said that it was less prohibitive. Forty-three percent said that it was less prohibitive, 14 percent said it was more prohibitive, and 43 percent said there was no change. For those that said it was less prohibitive, they emphasized the reasons why the barrier or the politic is coming down is that there’s greater awareness and emphasis on grid and understanding of it at the executive level. It’s a business decree: thou shalt do it. And they do it. And other parts of the thing companies are starting to understand is that using grid actually breaks down the barriers between the businesses and the silos and gives people more access, not less access, to the resources.
InfoWorld: What do you see as the major benefits of grid?
Baird: Grid offers a variety of significant benefits. On the bottom line profit enhancement, we’re talking about taking IT capital expenditure reductions and IT operational expenditure reductions. So you’re taking cost out of the equation. On the flip side, and offering a much bigger ROI, is the top-line revenue enhancement, and this is when you are using the grid technology to accelerate the time to discovery -- you are accelerating and improving business processes and operations, you are enhancing applications and the way that those applications perform such that the time to discovery, the time to result, the outputs are done in a more often systematic, effective, and obviously more shortened timeframe.
And if I take this a little further, with our clients we have done a deeper dive to understand the kind of net grid impacts that people are seeing on average. Typically we see with our clients, they’re telling us this, that they reduced hardware and software capital spending by between 20 and 40 percent after deploying a grid. That relates to the annual cost of the server, etc. They also reduce or reallocate an IT admin headcount fully loaded with compensation and everything in there of between five and 10 percent. And then they also reduce IT operation spend, the annual cost per server, as it relates to the support costs and the percent of the asset use and the plant and equipment cost reduction. And they are seeing between 20 and 30 percent typically after deploying grid technology. That comes from the bottom line.
You go to the top line then and the metrics change somewhat, and you’re talking about the ability to deploy a new application faster, so to get a system up and running and a new way of doing things in a shorter period of time. Then you go to things like increasing the job or work or transaction throughput, and this is the ability to do -- generate more income from additional work output or potential revenue from faster time to results or time to market. By increasing their work or job throughput or their transaction throughput, they get these significant gains. They are typically seeing between 10 times to 25 times, on average, increase in job or work or transaction throughput.
Another area that you see top-line ROI is the service-level agreement delivery. Typically, after deploying a grid, [users] are seeing that they’re losing virtually no jobs. Percent of unplanned down time is virtually nil. The quality of service and the percent of the quality of service increase has been going up significantly. Another area is increased IT resource utilization, so the actual IT resource and the utilization of that resource by the various parties that are drawing on it have gone up by 25 to 75 percent on average.
InfoWorld: What do you see as the technical obstacles grid faces?
Baird: There are a variety of obstacles for grid computing. There are technical capabilities and the technology is existing today to deliver grid computing without virtually any change. However, to get broader-scale widespread adoption of grid, there’s going to be additional work required. And one of those technical areas, or a couple of those technical areas, relate to standardization.
So grid is about inherently heterogeneous environments being integrated together, being consolidated, virtualized, and managed together. And therefore, if you’re talking about Linux connected to NT connected to Unix or IBM hardware with HP hardware with Sun and different applications, then standards are going to be key for how all this interoperability goes on. Standards related to security are going to be key as you move forward.
The next area is that not all applications today can work on a distributed computing environment. That means you need to be able to grid-enable these applications. Given that context, what we’re doing today and do with our customers on a regular basis is understand which applications they do see as targets for grid computing, and then we’re working with the vendors, and we’ve done it with some 100-plus vendors at this point, where we have grid-enabled their applications. That can be as simple as simply writing APIs or it may be more in-depth recoding and changing of the technology that is the software, in that case, to work on this middleware environment.
InfoWorld: What’s happening with standards in grid?
Baird: A lot is happening in the standards space. The central place for grid computing standards is the Global Grid Forum. And in that context then there is a flagship architecture known as the Open Grid Services Architecture, which will eventually be a whole set of specifications for different elements of grid technologies working together: Things like security, how do you deal with data over grid, etc. In the same vein then, one subset or one sub-specification of OGSA has been the Open Grid Services Infrastructure.
And the OGSI specification, which was developed about a year ago, has recently been re-factored to move it to a Web services-based environment, so it is now called the Web Services Resource Framework. And that is moving from a very state-based world that OGSI was to a Web services-based world in a simpler specification that others can work with to, in one area, help people grid-enable applications, amongst other things. And so that technology or that standard has recently come to the forefront only in the last few months, and it will now migrate from the Global Grid Forum over to the OASIS standards organization, which has a more Web services focus, and we’ll continue to track down that one elements of the overall open grid services architecture.
InfoWorld: How has grid made inroads in enterprise computing beyond its original use in scientific and research applications?
Baird: Grid has made huge inroads into the business side of the equation in that most of the original story around grid, as you pointed out, came from the research, academic, and the very high-performance computing side. What’s now happened in the last number of years is there is a lot of business computing work so you’re talking about analytical work, you’re talking about simulations and other work that is happening on the business side of the house that operate under similar types of attributes to the technical computing. So they often are large batch computationally intensive, resource-intensive work that’s happening. And so as we’ve moved from technical computing to business computing, there is almost a convergence of the needs of those two groups.
Where we’re seeing now is that the business computing environment is where the technology is starting to go and where people are starting to find application for it. But this comes back to the grid-enabling of apps and what people are asking for before those apps can be really used. So you see in the study that we also just completed, we asked people which were the most important apps that they wanted to see running on the grid, and in that case people are asking for database management systems, application development environments, BI and data warehousing, etc. at the top of the stack for applications they’d like to run on the grid. And as more of these vendors are putting these applications into a format that would allow them to work in a distributed computing environment or tap into the power and the capabilities of that grid world, it’s driving more and more acceptance of the grid into the enterprise space.
The other element that’s happening is that as awareness and understanding and more proof cases come out and people start to hear examples of J.P. Morgan/Chase with their platform technology saving millions of dollars by deploying it and getting far faster and more effective outputs, that drives it into the enterprise computing side in a very significant way.
InfoWorld: Do you see a market emerging for compute cycles generated by grid, similar to the electric power grid in which compute cycles would be sold off?
Baird: Yes. We actually look at grid and its ultimate development basically across sort of a three-step process. And the first stage we’re actively involved in at this point is what we call the enterprise grid, which is a grid inside the firewall. So inside a corporation, they’re sharing the compute resources across the corporation, across possibly large geographic boundaries, but still inside the corporation.
A partner grid then is the next phase which typically is grids across mutually trusting organizations. They likely set up a VPN or use certificate authority-type mechanisms to be able to secure themselves, but they’re sharing resources. That is a fairly early stage for the partner grids at this point because there’s a lot of standards and security issues and also just political issues that have to go on when you start going outside of an organization.
But the third level, and which we think is probably five to 10 years out at this point, is what we call the utility grid, the global grid, the service grid, etc. And this is where people start to really talk about utility computing in the sense that something analogous to the electrical grid where you’re going to be able to tap into the wall or into some network and be able to on demand request your compute resources, your storage resources, your memory, your applications, etc., and pay for that on a charge-back type of mechanism. Today that’s happening inside the enterprise, so you’re seeing some utility computing models happening inside the organization where different departments that have different P&Ls are actually tracking the use by other groups inside the organization of their compute resources and billing them back to the appropriate department.
As we move forward and as the technology and standards evolve, you will see a move towards utility centers like computing where maybe a telco, for example, could be providing the service mechanisms back to the end-user. So as today some telcos provide network services and VPNs and others on a per fee-use basis to their clients, it’s not a far extension to think of them as setting up datacenters or a compute resource environment, that a company could be linked via the networks that they’re already being provided by the telco to lease additional cycles, when perhaps they’re doing a job and they need more compute power. They could simply push their jobs down the pike to the datacenter offered by the telco, do the job, the work is sent back, and an accounting mechanism that we already have today in the enterprise grid space of platforms, technologies, etc. would actually bill back or charge back or record the amount of usage, and then a fee per that usage is billed to the customer.
InfoWorld: You mentioned telcos, and they’re of course large IT system users. What about a small company selling back cycles to some kind of grid consortium and then getting some kind of fee in return for allowing use of its compute resources in a grid?
Baird: Right. I think the possibility is there, the notion that many different smaller players that might have additional resources that could provide them [through] some sort of utility mechanism that is set up to accept additional resources and manage those resources. But I think that’s going further out into the world of the vision for grid. I think that there’s a lot of work and a lot of political sharing issues that would need to go on, and so I think it’s more likely to be a pretty straight line, contiguous kind of grid scenario where you are talking from one vendor back to another vendor that’s providing the additional cycles. Now that vendor, the telco in that case, may be providing those cycles to many different players at the same time, but I think it’s probably going to be a bit more centralized at least at the early stages, but what you’ve just set up though is absolutely within the realm of possibility.