Columns

Notes on the Grid: It’s a network problem

• By Jack Vaughan
• September 29, 2003

A few headline events this summer fell outside the usual realm of computer systems, but nonetheless triggered interest among mainstream technologists. Chief among these was the breakdown of the Northeast and Midwest power grids.

On August 14, some sets of circumstances, coupled with inadequate responses by power grid monitors, contributed to throw portions of the East and Midwest regions of the U.S. into a blackout. A cascading effect of tripped networked power plants -- that began, perhaps, in Ohio -- left air, phone and traffic systems in an uproar, and tourists sleeping on the summer sidewalks of New York. Years ago, in a different power system, the Ohio problems would have remained local.

The power grid’s bad hair day happened just as some large computer and software companies had begun to gleefully latch on to the idea of grid computing. A lot of work needs to be done before computer grids can move from a few scientific test beds to large-scale commercial deployment, but that has not slowed vendors from selling grid like hotcakes. Long-time industry hands have seen it happen before.

The power grid provided a quick analogy for the computer grid. In this vision computational jobs are moved about and handled as needed, much as electric power is pushed about the grid based on real-time requirements. Like most software analogies, the mapping between hype and reality is incomplete.

Such analogies can be useful for communicating concepts, but they can also be useful for creating half-truths for purposes of obfuscation. Grid has real possibilities for improved computing, but the sales pitches that come with it can overdo its imperviousness. Of course, even with a major meltdown, an electric power system still looks a lot more dependable than a computer system. Just as the power fizzled in the East, e-mail systems everywhere were breaking down as hackers fiddled with remote procedure call vulnerabilities in Microsoft Windows. The cascading effects of failures in power and e-mail systems seemed to make a musical sound like that of family TV pioneer Lawrence Welk’s old bubble machine run amok.

* * *

The power grid was a bit more suspect in the aftermath of the meltdown, but Oracle, which has been on its heels of late, nonetheless pushed ahead in early September with its planned grid marketing onslaught at OracleWorld in San Francisco.

Chortled company head Larry Ellison: “Grid computing is the biggest change in enterprise computing in 40 years.” Some resonance was probably lost in the wake of the big power grid breakdown. But there was no turning back; the PDF slides were already baked for the Moscone Center event. And Oracle 9i began to make way for Oracle 10g.

There is some steak here -- Oracle does have a somewhat formidable track record at balancing database clusters -- and some marketing sizzle -- Oracle rode its “Internet-ready” 9i release to new glory, and the firm can reasonably hope for some more glory on the trail of 10g. The connecting thread is the promise of the network to do something extraordinary.

* * *

Network effects are much discussed, but it is hard to put too much reliance on anyone who claims to understand them too deeply. Mathematicians and network engineers have been studying cascading effects and network effects for years, and there is no question that Claude Shannon at Bell Labs, Bernard Mandlebrot at IBM and others made significant progress in this realm. But one sees more and more on the topic, and wonders if theories are coming too easily.

Networks are ubiquitous -- meaning they can encompass most everything on earth if the definitions are pushed. Network theory is being applied these days to society, financial markets, ecospheres and soda pop marketing. When they get to soda pop marketing, you might want to put up a protective force field.

That the network can be more valuable than the sum of its parts -- or at least behave differently than a study of its parts would forecast -- is well known to software development teams. Add to that the well-known fact that strange behavior occurs when apps are placed on the network. This was a common problem in client/server computing nets, and it was never so much solved as jettisoned for another problem (Web-centric computing).

At the heart of grid, peer-to-peer and Web services is the notion that networked systems will create new markets for software but, as you who read the small print know, they will likely create new headaches, too.

In biology, where advances in identifying the elements of life have been slow to translate into medical cures, network theories may lead to approaches that will bring understanding of the organism as a whole, rather than as a collection of parts. That could be a very positive development.

But less obvious in its merit is networking theory as applied in terrorism prevention schemes such as Darpa’s Policy Analysis Market, which also made technology headlines this summer when U.S. Senators Ron Wyden and Byron Dorgan suggested that analytical data mining pools handling “gentlemen’s wagers” on the likelihood of assassinations or terrorist attacks were morally abhorrent. Behind the Policy Analysis Market concept was the idea that a network of informed individuals could forecast catastrophe.

Nothing promoted the power of the network more than the World Wide Web. It grew like a virus colony. Bob Metcalf came up with a law to describe it. Everyone could see this nonlinear burst in action, including the purveyors of Ponzi schemes, spam and worms.

Some of the latest thinking in network effects, the scale-free network, comes from observing the Web in action, where we are all nodes on a network, but some nodes are more highly linked to than others.

* * *

None has ridden the magical phenomena of networks further than Sun Microsystems. Through the 1980s and 1990s, the company was able to encapsulate the vitality of its Unix-flavored networking expertise with slightly varied sonorous sound bites: “The computer is the network” and “The network is the computer.” This was followed by the slogan: “We put the ‘dot’ in ‘dot.com.’” This all resonated.

However, the resonance is a little short in supply just now -- just when network effects are about to get big again. And one of Sun’s majordomos is leaving.

If, in the words of the old doo-wop song, you wanted to shake the hand of the man who put the dot in dot.com, to read recent press, you would head out to find Bill Joy. But the Sun co-founder and chief scientist is not holding down a formal job at Sun anymore. Of course, some outsiders would suggest, that’s not a big change -- his visits to the Sun plant were not so frequent, they assert, in recent years.

Joy in his heyday was certainly able to do extraordinary things with the Unix base product that came out of Bell Labs. And he played a major role in the development of the SPARC RISC chip, although, after meeting some of the brilliant chip designers that worked at his side, it was a little difficult to grant him the title of ‘Chip Architect’ that he sometimes seemed to covet. Maybe ‘Chip Champion’ was more appropriate. In recent years he has been seen on the Charlie Rose show somberly worrying about the threats of nanotechnology terrorists. With Sun’s future at stake, this somber worrying probably did not sit well with all those soldiering on to dig Sun out.

Sun’s Scott McNealy said the technology course of the company is in good hands with Greg Papadopoulos, the present CTO. So I headed with anticipation to the recent OMG Integrate 2003 conference in Boston to hear a Papadopoulos keynote.

A look at this reporter’s notebook shows that Papadopoulos can “talk the network talk.” “Networks have their own characteristics, especially as their use grows,” he said.

More from notes: At work in the future will be “automation of new levels of value creation” under the auspices of “distributed innovation.” We are entering an era where we will experience “the power of indirection.” In these coming days, “networks will decompose the PC.”

A considerate speaker, Papadopoulos engaged the audience with some bits of techno humor:

On Web services: “We [as an industry] are trying to find the moral equivalent of Unix pipes at the network level.”

“In the future, you will pay $59 per month for someone to take you off the network.”

And a comment on MSBlaster, et al.: “My son is wearing a T-shirt that says: ‘Your Windows system is my other computer.’”

This is good stuff. But Sun is definitely entering another phase where “walking the network walk” will be the important thing. Which way will McNealy and company go? It is time for some “what if.”

The Joy departure may be a clue. Partnerships and even mergers are often steps on comeback paths in high-tech. Joy had a major hand in the 1980s in Sun’s bet-the-company move from Motorola’s 68K to a Sun-designed SPARC microprocessor. What if Sun was ready to saddle up very close to Motorola, yet again, and Joy took a dim view of that? Or, worse, what if Sun was ready to get tight with AMD, which has endorsed the dreaded, less-elegant-than-SPARC Intel architecture that is eating away at Sun markets? What would Bill say?

The tab will come due for a next-generation chip design, one that Sun may not be able to afford. Hewlett-Packard admitted as much itself in the early 1990s by ceding chip development responsibility to Intel. Sun’s turn may come too, though few see Intel as the apt partner.

What's your take on network and related issues? Please let us know at Feedback