Subject: Re: Successful FSBs
From: "Tim O'Reilly" <>
Date: Sun, 27 Oct 2002 16:01:51 -0800

On 10/27/02 12:08 PM, "Ian Lance Taylor" <> wrote:

> "Tim O'Reilly" <> writes:
>> I've been saying for years that the shift towards commodity software
>> (whether free or just open standards) would lead towards a new paradigm in
>> which money was increasingly made on services.  (At one point I was calling
>> it infoware, now I'm saying web services and 'the internet operating
>> system', but the point is similar.  People don't pay for the software, but
>> for the services the software delivers.)
> I think that even ``web services'' is not the best name, as it implies
> a service provider and a service consumer.

I agree that it's not the best name.  "Web" is sort of irrelevant.  And of
course "web services" now implies SOAP et al, when in fact the use of CDDB
by CD burners or the use of DoubleClick by web advertisers is just as much a
"web service" as any SOAP service.
> Napster was a web service because it had a single point of contact.
> But in a properly implemented Napster, as some of the current music
> trading systems are, there is no single point of contact, and no
> identifiable service provider.
> Similarly eBay is a web service provider joining sellers and buyers,
> but in an age of infobots there is no need for a single trading floor.
> You just need agreed upon protocols, and you probably need to enable a
> market in buyer assurance programs.

I love the general thrust of your comments, i.e. to point out the various
distributed protocols as components of the internet OS, but I have to say
that I don't agree with you that "there is no single point of contact and no
identifiable service provider."

In the case of the music services, which have been designed specifically to
avoid the fate of Napster, there isn't now an identifiable single service
provider, but I think that this is a special case and a situation that won't

Once the music industry comes to their senses and offers a commercial
alternative on reasonable terms, I do think that there will be regular
points of contact, just as there are for current ISP-type services.  Yes, we
could all provide them to each other cooperatively if we wanted, but it's
easier to pay someone to provide reliable service.

I still remember the days of ISP peering, and the comment of one admin (I
think it was John Curran of NearNet, but it might have been Rick Adams or
Mike O'Dell over at Uunet) that:  "Right now, the small ISP is a peer, but
before long the big guys are going to say, "I know what you are.  You're a

And so, for example, the billing relationship and the convenient local POP
become the single point of contact for most of us in what are otherwise
fairly fungible IP services.

> For a current example, SMTP is not a ``web service.''  It is simply
> available on the Internet--part of, as you say, the ``Internet
> operating system.''  HTTP itself is not a ``web service,'' although it
> obviously enables them.

> I don't think we're going to eliminate the middle-man.  But I do think
> we're going to eventually eliminate the service provider.  The term
> ``web services,'' while it won't be wrong, will increasingly have the
> wrong connotations.  What the Internet does is bring interconnectivity
> to the masses.  I'm not good at names, but maybe this should be
> ``fully distributed services.''

Even though I've been a big champion of peer to peer computing (see, reading from the bottom, since articles are in
reverse chronological order), I also believe that it makes a lot more sense
for ad hoc groups a la Apple's rendezvous than it does for wide scale
applications.  Why?  Scale.

If you look at the actual behavior of peer-to-peer networks, they rapidly
get clumpy.  Gnutella quickly got super-peers, for instance.

But we don't even need to look at current p2p networks.  Just take the web,
which after all has the potential to be as decentralized as any network can
be.  (Any site can link to any other on a completely ad hoc basis.)  Yet in
the end, sites like google and yahoo! (or for geeks, sites like slashdot or
even have ended up emerging from the pack to become "more
equal than others."  Google's success is a testament to the emergent
dynamics of a flat linkspace.

The explanation was given in a different context by Jared Diamond, in Guns,
Germs and Steel (page 268), in explaining why improvements in food
production led to population centers, and then, to complex social
organization, as conflicts between unrelated strangers required structures
of intermediation:

"Relationships between a band of 20 people involve only 190 two person
interactions (20 people times 19 divided by 2), but a band of 2000 would
have 1,999,000 dyads."

Who can keep up with all the sites and all the mailing lists?
Intermediaries quickly emerge.

And back to music.  Networks like kazaa only solve the distribution problem,
since they rely on outside parties (the music industry) to market their
"product."  Without the artist/song being already known, you need other
means to find what you like.  Inevitably recommendation engines, search
engines, or other aggregation tools would be required.  Yes, you could have
bots that go out and find "more of what I like", but my guess is that bots
with access to more metadata and bigger databases (i.e. With access to some
degree of centralized services) will always be more successful than
completely decentralized ones.

If we had distributed shopping bots, I imagine that some bots would be
smarter than others, and it would be preferable to subscribe to the smart
bots service than the dumb bots service.  Especially with the scale problems
of searching a huge cyberspace, the site with the best concentration of
metadata and the best methodology for leveraging it (think google) wins.

P2P is an amazingly useful addition to the toolset, but it won't change the
fundamental dynamics of large systems.

For lots of discussion on this type of topic, see Lucas Gonze's great
mailing list,

> In a world of fully distributed services, profits for the service
> providers drop from the level of something like eBay to the level of
> something like a law firm.

Well, it's certainly true that most ISPs have relatively low profitability,
as they offer undifferentiated services, it's also clear that they failed to
take advantage of the opportunity to move up the value chain.  Microsoft
hasn't made that mistake, btw, nor AOL.  They realize that they need to
offer more than just the basic IP services if they want to capture more of
the revenue and profit, rather than leaving it to third parties.

There are lot of analogies to television.  Television started out "free" --
ad supported over the airwaves.  Yet now, almost everyone pays a monthly
subscription fee (cable).  Basic cable is one level, but then there are
various aggregated levels of premium channel.

But back to Ebay.  EBay's proprietary IP isn't their main competitive
advantage.  Even more than amazon or large ISPs, or cable companies, their
*customers* are their advantage.  In truly distributed networks, a la EBay,
the advantage goes even more strongly to the player who provides services
that make it easier for the independent parties to operate.

Tim O'Reilly @ O'Reilly & Associates, Inc.
1005 Gravenstein Highway North, Sebastopol, CA 95472