Deutsch's Fallacies, 10 Years After
In the fall of 1991, when mobile computing involved a hand truck and an extension cord, the idea of an everything-connected world was a leap of faith to some and a really crazy idea to most. But Sun's engineers were already working on notebook computers, and Peter Deutsch, one of Sun's original "Fellows," was heading up a task force to advise Sun on its mobile strategy.
Deutsch just called 'em like he saw 'em. When he got to Sun he began to consider some of the existing notions some engineers had about network computing, some of which were downright foolish.
Coming off a stint as chief scientist at Park Place Systems, Deutsch was looking to hang out and cogitate with and bask in Sun's intense engineering culture. He was a key designer and implementer of the Interlisp-D system and a significant contributor to the design of the Cedar Mesa language and the Smalltalk-80 programming environment. But he hadn't gotten into networking. It might have been some sort of intellectual hazing ritual that made Deutsch co-chair of a mobile computing task force. Or it might have been brilliance.
Bill Joy and Dave Lyon had already formulated a list of flawed assumptions about distributed computing that were guaranteed to cause problems down the road: the network is reliable; latency is zero; bandwidth is infinite; and the network is secure. James Gosling, Sun Fellow, Chief Technology Officer for Java Developer Platforms, and inventor of Java, had actually codified these four, calling them "The Fallacies of Networked Computing."
"It's a sort of funny thing," he says, "that in the large-scale world, networking didn't really exist in 1995. But since Sun was founded in 1982, networking has been at the core of what we do. We cut ourselves on all these problems pretty early on."
What Deutsch saw with fresh eyes was that, despite Gosling's warning, as engineers - inside and outside Sun - designed and built network infrastructure, they kept making the same mistakes, based largely on the same basic yet false assumptions about the nature of the network.
"The more I looked around at networking inside and outside Sun," Deutsch says, "the more I thought I could see instances where making these assumptions got people into trouble." For example, Deutsch could see Gosling's Fallacies coming into play as Sun moved its operations from downtown to its glamorous new campus near the San Francisco Bay.
"There was a lot of thrashing around about the topology of the network for the corporate intranet, where routers should be, etc.," Deutsch recalls. "Things broke all the time. My recollection is that it was watching all that thrashing around that led me to numbers five and six - that there's a single administrator and that the topology won't change." Number seven, that transport cost is zero, coalesced as Sun discussed creating a wide area network to connect the Mountain View campus with a new lab on the East coast.
When Deutsch wrote the list, by this time seven items strong, into a slide presentation, "It was no big deal," he says. Neither was there a roar of acclamation. Rob Gingell, Sun vice president, chief engineer, and Fellow, remembers it as an mmm-hmm moment rather than an ah-ha moment.
But putting it down on paper, codifying it, made all the difference. "The list Peter wrote down was a very pithy summary of the pitfalls people typically fall into," Gosling says. "We had many conversations about it. There would be a presentation where somebody would be proposing some design, and somebody would point out, ‘You know, that kind of depends on the network being reliable. And that's false.'"
Gosling added the Eighth Fallacy - the assumption that the network is homogeneous - in 1997 or so. "It reflected a mismatch between our perception and others' perceptions of the network. That the network is homogeneous is never a mistake we've made. But it was clear that lots of people on the outside had a tendency to fall into this."
If the Fallacies pervade networking, Sun was perhaps ideally placed to identify and surmount them, says Gosling. "We have a very strong engineering culture," he says. "A lot of our work is about building solid reliable systems."
Beginning with Java, Sun has explicitly grappled with the Fallacies. Java's "write once, run anywhere" approach celebrates heterogeneity, according to Gosling. While nobody exactly tacked the Fallacies up on the wall when Jini was designed, says Gingell, "If you look at Jini, it's apparent that what it's trying to do is confront some of the Fallacies, because the Jini team and Peter shared a view of those problems." Jini lets diverse devices discover and interact with each other without any administrator at all; it also can deal with multiple administrators acting divergently.
Indeed. "Almost everything in Jini is about dealing with that list," Gosling says. "It's all about the dynamic, spontaneous reconfiguration of networks in the face of a dynamic environment. Things getting plugged in, things getting broken."
The Liberty Project is another example of Sun engineering's acknowledgment of the single administrator fallacy, according to Juan Carlos Soto, product marketing group manager for Project JXTA and the jxta.org open source community manager. "Liberty provides a federated identity mechanism that makes it convenient for businesses to interact while still respecting privacy, because there's no single point of control," he says.
JXTA goes perhaps furthest in turning all of the Fallacies into truths. It creates an ad hoc virtual private network among peer devices that aren't always at stable addresses, Soto says, so it solves the problems of multiple administrators, changing topology, and heterogeneous networks. Moreover, because JXTA makes it easy to create multiple instances of a service, and peers cooperate to move messages through the network, it creates great resiliency that counteracts the inherent unreliability of networks. It's designed to be efficient and conserve bandwidth as much as possible, reducing the impact of Fallacies Two and Three, and it accepts every authentication method for flexible and tight security.
"P2P is turning the computer into both a server and a client," Soto says. "It's way beyond a single administrator and a homogenous network. It abstracts a lot of those issues away."
The Sun days are a long time ago for Deutsch, who now, as president of Aladdin, consults for technology and venture capital companies. He's a bit bemused that his short list of big goofs has become institutionalized as Deutsch's Fallacies. "If you had told me when I was at Sun that 10 years later, this one page of maxims was going to be one of the things I was best known for," he says, "I would have been floored."
Although networking has changed plenty since 1991, and some of the Fallacies, such as betting on a secure network, are more obvious, they are all still as applicable, Gingell says, and they continue to be driven into the hearts and minds of Sun's engineers. "Cultural reeducation is more important than any one product," he says. "If all our engineers understand networking at a very large scale…the products will take care of themselves."
About The Author
Ingrid Van Den Hoogen is the senior director of software strategy and marketing
for Sun Microsystems. She was the founder of the JXTA project at Sun, and a member
of the Jini team.