Grid ready to square mainframe circle

Utility architecture is hot now, but it owes a lot to the behemoths of the 1960s, says Richard Sharpe.

Computing, 15 Mar 2004

The 'm' word can be written in these pages again without fear of being met with a loud chorus of boos. The core values of the mainframe used as the base for applications in the 1960s and 1970s are now a cause not only for celebration but also for emulation.

The hottest topics on today's computer architecture agenda are grid and utility computing.

Behind these developments lie the core values of reliability, availability and security that made the mainframe king of its time. IBM had an inelegant acronym for this - RAS.

Many companies have benefited from hacking away at the reputation of the mainframe. At the time they were attacking the core business of IBM.

Few companies flourished more as a result of the demise of the mainframe and the slackening of IBM's grip on the technologies of the industry than Oracle.

Its founder, Larry Ellison, was an early and persistent proposer of first departmental-based computing and then the client-server model.

Ellison has always tried to move just in front of the wave. Some waves, such as the network computer, turned out to be mere ripples.

But others became a destructive and creative force reshaping the landscape of technology.

Ellison likes to nail his flag to the mast as his boat crests these waves - partly so that customers know what they are going to get but also, I suspect, to make sure the Oracle crews around the world pull the same ropes.

This time the flag is grid. The latest version of the Oracle core product, its relational database management system (RDMS), is 10g, the 'g' standing for grid.

Out goes version 9 with its lower case i standing for the internet; in comes grid.

Version 10 is as significant a move as Oracle has ever made, the company now says. It is as significant a move as the shift to version 7 in the late 1980s, which left the pack of RDMS competitors in its wake.

And Oracle is now willing to acknowledge the values of the mainframe.

"With grid we return to a recognition of the core values of the mainframe of the 1960s and 1970s," says Ken Jacobs, the Oracle vice president of product strategy in the server technologies division.

"Déjà vu is back to the future: we don't know where we are unless we know where we have come from," he acknowledges. Where we come from is a world of high reliability in a data centre employing mainframe computing with an efficient exploitation of central resources.

The complexities of managing those resources were masked from users. They hardly knew whether two or 2,000 users were online at the time, provided there was enough resource to go round. Each user seemed to get sole treatment from the mainframe hardware and software combination.

Resources were dynamically allocated to users depending on their need. Once used, those resources were put back into the common pool and immediately became available for other users.

Much of this power was available over networks - distributed networks. Users could, for example, dial up and run their application remotely without having to estimate the processor cycles they would need or their temporary and fixed storage requirements.

Services such as IP Sharp's APL service provided worldwide access to mainframe-class power for hordes of users.

Moving on from mainframe
The paradigm of computing moved on from mainframe-hosted to networked.

"The mainframe was the beginning of a series," says Jacobs: minicomputers in departments; PCs on desks; the client-server relationship linking them; and the internet providing the universal train to haul the bits around.

Jacobs has seen them all rise and then fall into their place. He has been at Oracle for 23 years of its existence. His employee number is 18: Ellison's is 1; and numbers 2 to 17 have left the company.

Jacobs is not the only voice interested in grid computing who also praises the mainframe.

"The mainframe is the father of this," says Dr John Manley, director of Utility Computing Research at HP Labs in Bristol.

"It was a very large machine with parallel jobs in which resources were allocated according to priority and cleared up afterwards," he says.

"We learned a lot about the isolation of one application and service from another, and advanced scheduling with the mainframe."

Oracle uses the term grid computing, while HP refers to utility computing. IBM says there is a subtle difference between grid and utility computing.

"Utility computing is about doing business, rather than about the technology," says Dr John Easton, grid architect at IBM UK.

He says the grid is one way of delivering utility computing, but not the only way. Manley describes the difference between the previous round of developments and those of today.

"With utility and grid computing we are making it more difficult," he observes.

"It is no longer a single device that we partition up. The power is now geographically dispersed; it is now about different sorts of resources; and there is a much higher degree of heterogeneity."

Here another piece of déjà vu can help us: a reflection on the distributed systems of the past.

"Extremely high-quality work has been done on distributed systems in the past 20 years," says Manley. "If all of it has not found an expression yet, we can go back and mine it."

To get to where we are we had to walk forward step by step in evolutions, not revolutions, he says.

InfoNet, ANSA, Corba, Brokers and the like were necessary parts of the groundwork.

But some steps were not in the right direction, as it transpired. Distributed computing environment, for example, was ahead of its time and few users "bought into it", says Easton.

Think back to 1976 and the unfolding scene of distributed systems. Listen to Earl C Joseph, then working for Univac before it emerged as Unisys, and then a futurist, as he still likes to call himself.

He told the Infotech State of the Art Conference on Distributed Systems in 1976: "The near-term expectation is for computers to be distributed into everyday uses like household appliances, automobiles, offices and schools.

"Soon thereafter, we can expect their connectivity into computerised communications networks - common carriers and dedicated. The result will obviously be the further computerisation of society."

Joseph traces the development of the grid/utility model through 30 important steps in the development of computing, from multiple computers to timesharing and, most recently, clusters of computers.

As a good futurologist, Joseph has projected the line onwards. His reference to connectivity, and the type of service provided by the common carriers of the telephone world, indicates the emergence of grid and utility computing. Others made similar comments at the time.

Grid and utility computing, as Manley says and others echo, should be like the telephone network's dial tone: always there, global, hugely reliable and for your use for as long as you want to use it and can pay for it.

But countless IT projects in the past have failed because the newest technology was not robust enough to support them.

If déjà vu teaches us anything it is that each new computer architecture is not a universal panacea: it has a sweet spot that can deliver sizeable advantages. But in a wider context its success is questionable if not impossible.

Looking for the sweet spot
What does déjà vu tell us is the likely sweet spot of the grid/utility model? HP claims it is in applications such as computer graphics rendering, which demand huge amounts of batch processing power.

This is the current focus of Manley's work. He wants to provide a true utility service for animation houses, which cannot afford to pay for the permanent supply of the computing resources they need occasionally to render their images.

HP Labs has started this service as a pilot and rendered short animations with glass, water, smoke and dust that have proven the concept.

In the future Manley plans to relax the constraints in three or four phases to include multiple users. By early 2006, studios should be able to enter major film festivals with animations rendered through this service.

This rendering application is the type of batch application that best fits the grid/utility model, he says.

Later, as déjà vu shows, it will spread out to long-lived applications such as database resources.

Manley points to the 1901 census access as an example. The web site providing access to the digital images of the 1901 census was designed for 10 million hits. It was so popular that it attracted more than 30 million, bringing it down for everybody.

Grid/utility computing could have solved that problem by pouring in more processor cycles and storage. "We have to bring in resources and take them out according to load," explains Manley.

The commercial applications of the grid, he says, lean towards these long-lived applications. It's easier than that, proclaims Oracle.

Yes, in hindsight, many of those earlier transitions between computer architectures were difficult. But this one will be easy. Or easier.

"Of all the steps, the step to grid is the easiest and the most valuable move an IT department can make. Applications will not have to be significantly rewritten."

The systems software, not the application, should take care of the grid nature of the service. "There is no particular sweet spot for new applications," says Jacobs.

While Manley predicts it will take five to 10 years to achieve real exploitation of the grid/utility model, Jacobs and Oracle are far more go-ahead.

But Manley advises caution, and counters those who push the idea further than it can go today. "Outselling the promise of the grid is too dangerous," he says.

Oracle envisages grid computing will take the form of clusters of data centres providing services in a corporation.

Grid is as much a way of creating efficiency in the exploitation of existing resources as anything else.

The sweet spot for users that have large SMP systems costing millions is that grid computing could provide an application for hundreds of thousands, he says.

Among its customers for the technology, Oracle lists the Ordnance Survey and scientific research organisation Cern.

At times it seems that Oracle's vision of the grid is just the exploitation of a cluster of Linux boxes.

About a third of users simply buy more resources when applications seem to run out of capacity, according to research sponsored by Oracle in Europe.

Yet more than half of the users have no idea of the level of strain on their technology infrastructure. They buy more capacity as a knee-jerk reaction to a problem instead of seeing what they have.

"Use what you have better" has been a long-standing approach of IBM's. It now has 150 customers actively considering, working on or completing grid applications.

The key to IBM: standards. Like HP, it uses the Globes standards. Yet standards imposed too early freeze out innovation, as déjà vu teaches us.

How we got to the grid ...