You need grid computing. It could save you millions. It could provide competitive advantage to your business. But to get it, you have to build it yourself. Why? Ask your vendors.
So in 2003, Scott McKay, CIO and senior vice president of operations for Genworth Financial, tried something that few CIOs in industries outside of academia or financial services (where the need for massively parallel computing trumps all other considerations) have been willing to take a risk on until now: he adapted his custom application to work on a grid -- a vast, fluid pool of processing power that splashes around in a motley assortment of spare memory from PCs and servers throughout his company. McKay's shark never runs out of food on this diet, and he now processes actuarial tables in 20 minutes instead of five hours.
McKay loves grid and the money he saves, but he also knows why most of his CIO colleagues are still feeding their sharks the old-fashioned way: application software vendors won't let their sharks eat grid. That means CIOs who want to save millions on their infrastructures have to build grid applications themselves -- no mean feat. "There are a number of issues to make grid work," says McKay.
And the primary issue is (no surprise) money.
Sacred cash cow
Grid blows up the traditional software licensing model, which charges customers according to the computer processor that runs the application -- in other words, one application, one computer, one price. But with grid, no CPUs are dedicated to a specific application. McKay's grid application draws on a river of power that may flow across as many as 300 of his PCs in a day, but may only spend a few minutes on each. To date, vendors haven't been able -- or, say some critics, willing -- to figure out how to make their money with grid computing while giving users what they want -- virtually unlimited CPU power without astronomical licensing costs.
"The vendors haven't come to grips with how they're going to license their software," says Jonathan Eunice, principal analyst with IT research company Illuminata. "Software licensing remains messy and wholly unsolved."
At this point, vendors are unwilling to relax their traditional per-CPU software licensing models to allow grid to flourish, according to a report from The 451 Group, a research company. "None of the vendors are doing anything material to support changes [to the software licensing model] at this time," says William Fellows, principal analyst at The 451 Group.
The landscape is further cluttered by the conflicting claims that vendors make about their grid offerings in this hot, new market, says Carl Claunch, a vice president at Gartner Research. He says grid computing has been "hijacked by the marketing folks." And even some vendors agree (about other vendors): "What all the vendors have done is to put grid under their umbrellas, [regardless of whether their solutions really qualify]," says Ken King, IBM's vice president of grid computing. "That has created hype around grid, and confusion."
Other barriers to the creation of a solid grid market include a lack of standards and the high cost of reconfiguring standard client/server applications to work on grid.
But despite all these problems, the grid buzz refuses to die, because the potential for saving money, reducing complexity (for CIOs, anyway) and lowering maintenance burdens is far too compelling. McKay says he has slashed his hardware acquisition costs, chucked excess software and provided processing times that beat any of his competitors in the insurance industry.
And when an insurance company can process its decisions faster than competitors, you're not talking IT anymore, says McKay, you're talking competitive advantage. "Grid is one of things that differentiates us" as a business, not simply an IT shop, he says proudly.
But vendors won't vote themselves a reduction in revenue anytime soon, says The 451 Group's Fellows, which means more CIOs will be left scratching their heads, wondering, Why not me?
The meter is running amok
For CIOs not steeped in the nuances of grid computing, cutting through the hype can seem daunting and not worth the trouble. Most of what vendors are offering is not grid. The list of grid pretenders is long, but it includes clustering, utility computing and virtualisation.
Once CIOs figure out whether a grid-computing infrastructure is right for one or more of their software applications, then they have to decide how much help is really available from vendors and how much they want to pay.
Finally, they have to be prepared to hurl themselves against the wall of software licensing. "The software-licensing model is the biggest impediment to grid computing," says Gartner Research's Claunch. "The vendors know they've got problems."
The few vendors who have tried to come up with a formula for grid pricing have ladled on the complexity, according to analysts. "If you look at the formal price sheets, it is a nightmare for the customer," Illuminata's Eunice says. "You can't easily buy software that is intelligently licensed for the kind of dynamic infrastructure in grid."
Yet, alternative licensing models have been suggested, according to Fellows -- such as one that would charge for actual usage of the application, not just for the absolute number of CPUs available in the grid. But these new models require vendors to embrace new forms of metering and billing mechanisms to track customer usage by time, users per month or transaction volume. Few mechanisms exist to measure the kind of application and component usage and licence tracking that grid implies, although some are being tested.
Even if those applications prove to be successful, there needs to be a huge leap of faith between the vendor and customer. Customers have to believe that they won't be charged for time they aren't using, and vendors have to believe that their customers aren't somehow sneaking CPUs or users past monitoring tools. A few big companies with big vendor clout have been able to negotiate individual pricing and monitoring models, says Illuminata's Eunice, "but for medium and small companies, it's still too early."
Fellows says that CIOs he has spoken with aren't asking their vendors to reinvent the wheel yet. They would be content with some modifications to the time-based usage models that they already have. "If their [vendor] made some incremental changes to licence models, real simple things like rolling over minutes not used to a later time or not counting minutes used to review execution results," that would be a start, Fellows says. "They just want some tweaks to what they are able to do now."
Though no vendor has publicly stated its intentions to alter its pricing model for grid computing (at least not in its purest form), a few vendors have taken what Fellows terms "baby steps." SAP, for example, has been testing the open-source Globus Toolkit to grid-enable its products, which, if successful, might at least lead to pricing-model discussions. All it could take to liberate huge market share from competitors, says Fellows, would be for one powerful software vendor to simply change its model.
Certain applications that have been developed in-house have no licensing costs, such as analytic, financial calculation and electronic design automation. Those have been on early adopter companies' grids for years. But outside large corporations, analysts say there's a lack of in-house grid talent to make those in-house solutions work.
The grid horizon keeps moving
There are no explicit technical barriers to putting an application on a grid, according to Fellows. "If you really want to, you can grid-enable anything," he says. But if there is a legitimate argument to be made against a wholesale shift to grid, it is that converting existing applications requires lots of hard work, money and experienced grid developers (who are in short supply today).
Unless an application consumes a lot of processing power, CIOs might not see a big payback, and vendors might be left holding the bag for an expensive rewrite. The financial uncertainty surrounding software conversion costs and licensing models means that early adopters will probably pay a premium for grid-enabled applications, according to The 451 Group report, further slowing grid adoption.
So few applications have been rewritten for grid today (mostly analytical software) that CIOs are left with stark choices: bug your software vendor; convert an existing application to grid yourself, as McKay did; or build a grid application from scratch.
Here's why converting an application to grid takes a lot of work: The essence of grid computing is the availability of computing resources from all kinds of PCs and servers inside and outside your four walls. One attribute of early grid computing applications is that one task doesn't depend on the outcome of another task. (In big, number-crunching grid applications, for example, calculations are parsed into small, independent slices and can be added together at any time.) But many enterprise applications have dependencies -- one calculation or process can't move forward until another finishes. Developers have to figure out how to divide application processing into pieces (called threads of execution) to allow those applications to be run in parallel, according to Gartner Research's Claunch. Unless the code can divide its operations across dozens, hundreds, thousands or more threads, says Claunch, it will not be able to scale its performance when given the many computing resources that grid offers. Parsing the application takes work and requires a lot more than simply tweaking existing applications; it may even require a complete rewrite.
For all those reasons, CIOs haven't yet implemented traditional enterprise applications -- including ERP and CRM -- on grid, because in these, each task typically depends on the outcome of others, which is not something grids are good at yet, though any application can theoretically be grid-enabled. Forcing those applications, as written, onto a grid would be similar to strapping an Atlas rocket to a Volkswagen. "Your car would melt," says Illuminata's Eunice. The core algorithms inside many of these transactional applications would also need to change.
Observers aren't optimistic about the economic incentives for vendors to rebuild their wares for grid computing. "Grid will proceed at the pace with which the software changes," says Forrester Research principal analyst Frank Gillett. "Glaciers move faster."
Let's pretend to agree
Standard methods for gridding applications would help speed the glacial pace of grid acceptance by reducing the cost and complexity of the development process. Though standards bodies and alliances have emerged -- such as the Global Grid Forum, Enterprise Grid Alliance and the Globus Alliance -- competing grid definitions, proprietary technologies and entrenched vendor allegiances have caused them to collapse into warring factions, according to grid analysts. "Right now, we're at the end of the first [stage] where everyone's battling," says Illuminata's Eunice. "It's not the prettiest of pictures."
There are two major steps left to go before grid becomes enterprise-ready: First, the different groups must agree upon a set of standards; second, vendors need to build those standards into their products.
Fellows is not convinced the fighting will stop long enough to move beyond the first stage. "When users see more and more [standards] organisations, then users see more and more complexity," he says.
But open source may break the gridlock on grid. The Globus Toolkit (now in its fourth version) is free for anyone to use. Yet, the toolkit is limited; it's good enough to be effective at some tasks -- such as analytic and electronic design automation applications -- but not good enough to be effective at everything, says Eunice.
As for CIO uptake -- who knows? The Globus Alliance, the non-profit group that offers the toolkit, does little marketing and does not diligently track enterprise usage. Worse, grid skills are in short supply, which makes many CIOs unwilling to share project details, leaving neophyte gridders with no road maps to follow.
For all its early promise and economic benefits, in the end, grid computing may end up being a mirage for CIOs. They will continue to hear about how grid computing is an idyllic solution that everyone will be using in 10 years or 20 years (which they have already been saying for 10 years). Right now, says Gartner Research's Claunch, "CIOs can't see how to get from here to there."
For the few who can, such as McKay, the view is pretty good. It's taken his IT department more than two years to get to where they are now with grid computing, but it's been well worth it. "Our ability to execute more effectively, to achieve the business results, is huge."
Too bad most CIOs can't share those results.