[Economist-TQ] Thoughts on decentralization

[Rohit Khare]

To the Editor --

I would like to congratulate the Economist's computing correspondents 
for another fine job summarizing the piles of trade rags on my desk 
into a few pages with the latest TQ.

One sentence that caught me, though, was in the Sun piece: "The 
concept of distributed computing -- ie, peer-to-peer [P2P] 
networking..." Rarely have I been so peeved by an "ie"!

To steal a phrase, what is effective about P2P is not original, and, 
so far, what is original about P2P is not effective. We have long 
known how to merely break up a problem and distribute it to multiple 
workers _within_ a firm. The central challenge of the age, both in 
business and technology, is decentralization, not distribution.

Here's another example of misapplying that lesson, from the same 
article: "All those who use an instant-messaging service... are 
communicating directly with one another via a peer-to-peer network 
with almost no central control whatsoever. Likewise... Napster"

In fact, the bitter politicial feuds around IM and filesharing are 
direct consequences of *not* decentralizing. Millions of users of AOL 
Instant Messenger (AIM) can only communicate with other AIM users -- 
and if they think of switching, AOL's exclusive control of their 
Screen Name gives pause. (To be sure, the other centralized IM 
services are as guilty.) The centralization of Napster's databases is 
precisely what made them victims of legal challenge, too.

How do you compute when all the other computers are very far away and 
owned by other firms? That's what the best of "P2P" and "Web 
Services" research is essentially about. Even, for that matter, the 
challenge that the EC's Gallileo constellation poses to American 
control of GPS (which, surprisingly, was not mentioned at all!).

I look forward to the Economist's continuing incisive coverage of 
this central architectural issue in the future...

Sincerely,
   Rohit Khare
   Founder & CTO
   KnowNow Inc.

=========================================================
[FoRK] Thoughts about 'decentralization' provoked by the latest TQ:

-- Sun's N1 and JXTA: Turning "the network is the computer" inside out??

It is amusing that Sun continues to transform its image into the 
grumpy old uncle of the computing business, merely by scoffing at 
every new trend as a pale echo of what their founders already 
mastered twenty-five years ago.

Even if it is true. :-)

"Unix... was designed to manage separate pieces of hardware as if 
they were all part of a single computer." That's as concise a 
definition as I've ever heard of the fallacy of distributed 
computing. As the expanding universe of computing devices shatters, 
the kingsmen of distributed computing run around vainly trying to put 
Humpty-Dumpty back again.

"a stack of protocols that will allow all computers to find and 
connect to one another without any form of centralised help." That's 
also as concise a definition as I've ever heard of the essential 
challenge of decentralized computing: coming to agreement without 
appeal to an external arbiter. Whether it's as neutral as IEEE's 
stewardship of Ethernet MAC addresses, or as charged as ICANN's 
control of domain names, we always outgrow centralized arbiters, 
however light their touch.

"'Grassroots movements aren't easy to organize,' says Mr. Joy." 
Ironically, as true of the JXTA community as JXTA itself.

-- Fuel Cells

Again, this is a story of decentralization. In this case, of moving 
the means of (power) production into the hands of the consumer. But 
the interesting part of the problem is not "distributed generation," 
as the concluding paragraph puts it -- not merely moving the parts of 
the grid around physically, but shifting ownership of the grid to 
many new agents.

I suspect the next front in the struggle for decentralized power will 
shift from funding chemistry research to the protocols, pricing 
regimes, and interconnection standards for allowing "foreign" 
agencies to join the incumbent grids.

[Two local notes: Kudos to our own UC Irvine NCFRC for the lead on 
this piece. And isn't it amazing to live in a megalopolis with 
undreamt of infrastructure already underfoot, such as LA's hydrogen 
pipelines?]

-- Powerline networking

The Neverwire -- what a beautiful paradox of naming! -- is another 
hint at the importance of decentralization. It is not merely enough 
to distribute signals across powerlines; the hard part is ad-hoc 
construction of secure enclaves that decentralized one household's 
traffic from others on the same grid.

-- Spider silk

Sometimes, of course, centralization is much to be desired : "The 
problem is that spiders are not silkworms, which can survive 
peacefully in close quarters... By contrast, spiders are too 
aggressive and territorial to domesticate. 'It's like farming tigers 
-- they eat each other,' explains Jeffery Turner, boss of Nexia 
Biotechnologies"

-- Satellite radio

This is quite an upside-down decentralization parable! By physically 
centralizing broadcasting, sat radio promises to decentralize 
programming. The thousands of terrestrial analog broadcasters are 
proud of the fact that they deliver precisely 51 kinds of 
programming; the singleton sat companies are promising hundreds.

IBiquity's In-Band On-Channel (IBOC) simulcasting technique (packing 
digital data in FM sideband lobes) is another example of a basic 
Internet decentralization principle: backwards compatibility. The 
same  transmitted signal can be interpreted by legacy and future 
receivers. However, it does not exhibit *forward* compatibility: the 
added digital bitstream was not forseen by the analog format, nor 
does it appear to enhance the original signal -- only replaces it 
wholesale.

-- Satellite positioning

This is a great 2-page profile of the history of satnav, "an example 
of a self-perpetuating innovation: the better it gets, the more uses 
people find for it."

Its accuracy in positioning is a mathematical consequence of its 
accuracy as a clock, and it's a damned centralized one. "But the 
best-kept secret about GPS is its value as a clock. Time-sensitive 
operations associated with banking, telecommunications, and even 
power grids have come to rely upon GPS."

It is therefore instructive to consider what the consequences of all 
systems on the planet deriving the legitimacy of their clocks from a 
single agency (the US Navy). It works well when it does, but if the 
military re-activates selective availability -- or merely hardware 
failure of the satellites -- who is liable when ships drift off 
course, or stock trades can't be reconciled correctly?

Arguably, the only possible advantages of the Europeans' separate 
Gallileo satnav initiative over just waiting for US GPS 3 are related 
to decentralization. As a civilian alternative to complement GPS, the 
EC's architects have spent vastly more time thinking about the 
liability of centralized time and quality of service. The white 
papers for Gallileo propose a much more nuanced method for certifying 
clocks, and even decertifying rogue satellites -- as well as 
incorporating the sorts of privately-run, terrestrial differential 
beacons inevitably necessary for fine-positioning within the same 
trust management infrastructure.

-- AI, the return of

To actually cope with decentralization in the Semantic Web, we may 
well find ourselves right back in the AI textbook stacks. It is a lot 
more work to prove theorems like "is Roy's Drywall equivalent to 
Henrik's GypsumBoard?" than to try quick lexical comparisons -- and, 
ultimately unavoidable.

I'll hazard the generalization that AI has thought far more about the 
challenge of diverse agency than Software Engineering.

-- Machine Translation

A great example of the risks of hiding agency distinctions is in the 
quest for shared translation memories: "Over the past few months, 
however, a number of firms producing translation-memory software have 
suggested that translators might wish to pool their work... A bigger 
problem is ownership. Mr. Hutchins points out that translators may be 
unwilling to give away their work... Another problem when translating 
technical manuals is that many firms have their own proprietary 
terminology, which they use to distinguish themselves from their 
competitors."

Just imagine the same problem embedded as an exchange orchestrating 
auctions from multiple vendors' catalogs, and it is clearer how Web 
Services Integration faces the same problems in translating languages.

[Local note: had no idea Systran (of Babelfish fame) had Caltech roots...]

-- Foveon's X3 chip

[local note: I've been hearing about Caltech prof Mead's new chip for 
some time now, in general and alumni pubs. I'd never gotten as clear 
an explanation as I got in this article. Namely, that they separate 
color sensors in Z, rather than XY, by hacking the color-dependent 
penetration of silicon.]

-- Rick Rashid

Highly recommended profile!