OCEAN
The Open Computation Exchange & Arbitration
Network
An open platform and commodities market for distributed
computation
Michael P. Frank
University of Florida
CISE Department
mpf@cise.ufl.edu
Business Plan
(Preliminary Sketch,
11/16/00)
A.
The Business Concept, and the Business Itself
In one sentence, the concept is to found the
world's dominant commodities market for the buying and selling of active raw
resources of distributed computation (processor cycles, memory, storage, and
network bandwidth).
The OCEAN corporation would rapidly expand the
latent demand for such a commodity exchange by developing and giving away open
standards & software that will transform geographically-fixed computational
resources into a real-time, dynamically tradable, remotely-usable commodity,
similar to the market for electric power that is active today in much of the
US. An aggressive branding and
standardization effort associated with our distributed computing platform technology
(like Sun & Java) would establish the OCEAN brand name (or perhaps another
name, chosen more carefully) in the minds of the world as the central market for tradable distributed computation, in analogy
with the brand recognition enjoyed by NYSE or NASDAQ today. The capacity of the exchange (the
"arbitration network") could be scaled very rapidly by licensing the
core market-operation software to partner organizations under co-branding,
cross-market trading, and profit-sharing agreements, similar to the way that
existing stock & commodities markets partner with each other. Modest transaction fees, extracted from
bid-ask price spreads and shared among the partners, would guarantee the OCEAN
brand and corporation a steady income representing a small but meaningful
fraction of the world's ever-growing expenditures on computing power for the
foreseeable future.
B.
The Opportunity and Strategy
The
Opportunity: Applications as
diverse as bioinformatics, CGI rendering, nuclear explosion simulations, and
genetic algorithms demand heavy-duty multiprocessor computing power that costs
tens to hundreds of millions of dollars to obtain (cf. Wired, Dec. 2000, p. 340). Meanwhile, the collective computational
power of machines connected to the Internet is sorely under-utilized. We believe that therefore a market exists
for one or more commodity exchanges for
computation, which would permit the automated buying and selling (on a
virtual trading floor) of present and future reservations for remote use of
computational resources.
Technical
Strategy: The various types of resources and
reservations would be characterized and made tradable via an open, standardized
XML-based description language. The
usefulness of the traded resources would be enhanced via a set of open
technologies and conventions providing a universal distributed computing
platform, based on related existing technologies such as Java and CORBA, that
facilitates the use of purchased resources to run an end-buyer's particular
distributed computing applications. The
novel feature of the OCEAN platform will be that it give an application
designer the ability to have a given application to dynamically and
automatically (without need for per-transaction human intervention) purchase as
much or as little distributed power as is needed for a particular job instance,
and ship the job's executable code and data objects portably to remote
computation servers which can be provided by anyone. The supply-demand economics of the system and the commoditization
of the resources (including futures) will guarantee that on-demand
computational resources are plentiful at market prices, and that utilization of
resources will be high at any moment.
Growth
Strategy: OCEAN's growth involves three parts: (1)
Growth in the number of computation servers, (2) growth in the number of
OCEAN-based applications & the level of their use, and (3) growth (scaling)
of the computation-market infrastructure itself (the "computation exchange
& arbitration network" proper).
To address these:
(1.) The OCEAN software that allows a
computation provider to deploy a machine or cluster of machines (with their
associated memory, disks, and interconnections) as publicly-available
computation servers will be made freely available, so that anyone can download
and deploy a paying computation service at their site. This will enable rapid growth of the OCEAN's
capacity, similarly to what is seen in peer-oriented systems such as SETI@Home
or Napster: Any college student can set up a paying OCEAN server on their dorm
room computer, and any small business owning a cluster of machines can
configure their cluster as an OCEAN server when they are not otherwise being
used, thereby offsetting the cost of ownership of their computing
facilities.
(2.) The flip side of the growth coin will be
the growth of the installed base of OCEAN-based distributed applications and
their users. The expensive
supercomputing installations mentioned above demonstrate that the demand for
massively parallel computing power exists.
The OCEAN distributed computing platform should be made as compatible as
possible with existing parallel and distributed computing technologies such as
CORBA, JavaSpaces, etc., so that existing applications can be ported to OCEAN
with maximum ease.
But much more importantly, we foresee that the
publicly-available and general-purpose nature of the OCEAN infrastructure, in
contrast with all previous distributed computing platforms, will encourage the
rapid and widespread emergence of a latent demand for this new type of
computing platform, as software creators everywhere devise an enormous variety
of new applications for this type of computing power, whose scope we cannot
today hope to foresee.
For example, any
present-day computer application that from time to time is seen to run too
slowly is a potential candidate for conversion to an OCEAN-based version that
enables the application's user, if he or she becomes frustrated with the tempo
of the program's progress, to authorize the software to purchase (on the user's
behalf) additional resources from the OCEAN as needed in order to accomplish
its task more quickly (up to some maximum cost specified by the user). Whole industries will likely spring up to
provide software development tools that make it easy to write applications that
provide an "OCEAN boost" option - or such software can be developed
by the OCEAN corporation itself.
Or as another example, it could become standard
for every e-business to solve its scaling issues by writing its software from
the start to harness as many distributed OCEAN resources as is necessary to
handle whatever level of usage the business is currently experiencing. E-businesses could become completely
divorced from any particular hosting facility, and instead consist of a suite
of programs and data that migrate around the OCEAN, growing (or shrinking) as
needed.
But, thinking only of enhancing existing
applications is incredibly short-sighted: our guess is that an OCEAN-like
system makes computation a fundamentally and qualitatively different kind of
resource, being substantially more “liquid”, if you will, than previously; and
that as a result if OCEAN becomes widespread, it would spawn the development of
whole new classes of computing applications that would not have been
possible at all before OCEAN, i.e., new killer apps which we cannot
presently imagine and which are slated for some future entrepreneur to
discover.
(3) The arbitration network would be scaled by
distributing it as well, in a distributed network that nevertheless maintains a
single, consistent virtual trading floor by using well-known atomic transaction
protocols designed for distributed systems.
In addition, multiple separate trading floors, perhaps specializing in
different types of resources, could be allowed to flourish, though with interactions
between them, similar to today’s financial instrument markets. As stated earlier, licensing the core
technology to other organizations under modest terms would facilitate the rapid
growth of this multi-market structure.
The use of XML technology as the core description language for resources
and contracts ensures that the structure of these interacting markets could be
made as rich and flexible as needed to accommodate new innovations in the
development of these markets.
C.
The Target Market and Projections
The target market, as mentioned above, is
potentially every application running on every internet-connected computer,
which in the increasingly-interconnected future will mean every computer,
period. OCEAN will enhance the essence
of what computation is, and
computation itself in turn enhances every significant area of economic
endeavor.
In one of Gordon Moore's papers discussing the
Law that bears his name, he points out an interesting corollary to Moore's Law,
which is that if the computer industry as a whole continues its history of
exponential growth (in concert with the increasing performance/price of
computing), then by 2050 the industry would exceed 100% of the (historically
slower-growing) projected GWP (gross world product). Moore's conclusion is that the computing industry's long-term
growth will slow down before then.
However, we would like to suggest that
computation is a fundamentally different kind of industry than say cars or
clothes or airplanes: Increasing rates
of computation enable increased rates of improvement in every facet of the economy, insofar as the efficiency of any
industry or research effort is improved by the availability of more rapid,
inexpensive and far-reaching information-processing power. Therefore, as computation consumes more and
more of GWP, we consider it more likely than Moore's hypothesis that what will
actually happen is that as the two curves converge, the most important resource
and most of the world's economy will be
computing itself, and the rate of GWP
growth will increase to match the growth of the computing industry, at
least until the fundamental physical limits of computation are reached - limits
which are, incidentally, this author's major academic research specialty.
In this (admittedly somewhat optimistic)
scenario, we can project the growth rate of the total economic value of all the
world's computing power (which has actually been faster than exponential in recent decades, doubling in time periods
that get shorter each decade), and we can project it many decades into the
future to the point where we expect the fundamental limits reside (although we
cannot be sure of this point yet).
We project that the OCEAN, if it is developed as
an enterprise in the best possible way, could potentially carry a majority of
the trading volume of the shares of this computational capacity, drawing a
small fee from each trade. In this way,
the growth of the company's value can grow in concert with the fast-growing computational
economy virtually indefinitely, at least for many decades, until some future
world-transforming concept overshadows it.
Of course, the yearly value of the world’s
computation is already at least in the tens of billions of dollars; if just 10%
this migrates to harness OCEAN, then even only a 1% transaction fee would mean
tens of millions in annual revenue, and growing. And further, the widespread use of OCEAN would itself probably
accelerate the rate of growth of the computing industry.
D.
The Competitive Advantages
Other distributed-computation-for-pay ventures
exist, such as PopularPower.com and ProcessTree.com, but we do not see any of
the existing ventures pursuing an aggressive open-systems strategy geared
towards the rapid growth of the market, such as the strategy we envision. Thus, OCEAN has the potential to become the
first such market to be used on a massive scale.
As OCEAN grows, its competitive advantages would
rest most importantly on branding: on the trust and sense of security people
will feel when they rely on the OCEAN-branded markets as opposed to some
similar new competitor. To build the
brand, the OCEAN venture will need to clearly display its branding within its
freely available software and services at every possible opportunity, as well
as advertising it extensively in various media.
A second possible source of competitive
advantage is trade secrecy in the implementation of the core code of the OCEAN
"arbitrator" servers, which implement the market, store prices, and
match up buyers & sellers (as in a modern electronic stock or commodity
exchange). However, this would be a
weaker advantage, as the OCEAN functionality will not be rocket science to
replicate, once people see it in action.
Patenting is potentially a more powerful recourse, but patents are
subject to various weaknesses such as prior-art challenges as well.
For these reasons, the OCEAN venture will
primarily try to stave off competitive attacks by offering would-be competitors
reasonably-priced access to its technology under licensing agreements that
include profit-sharing and co-branding arrangements, as has been common with
many web-site technologies (e.g., search engines, mapping
services). A potential competitor
should perceive it to be obviously more painless to license the OCEAN code and
pay a small branding and royalty tithe than to develop a complete system from
scratch. Rapid initial growth of the
OCEAN standards due to the free distribution of its distributed computing
platform software will ensure that any potential competitor must be compatible
with the de facto standards set by OCEAN, so that any competitor will be
playing catch-up if they try to re-implement the core arbitrator software
without a licensing agreement. The
strategy is similar to Microsoft's strategy with the Windows platform where
they allow vendors to customize and co-brand it, while Windows imitations are
left behind. However, by adopting
modest and fair pricing strategies, OCEAN would avoid garnering ill will and
the legal challenges that arose against Microsoft's monopoly.
E.
The Economics, Profitability, and Harvest Potential
[To be expanded upon later.] The economics seem promising. The cost of operation of OCEAN's core
market-operation servers (“arbitrators”) will be modest: These servers will only
need to exchange a relatively small amount of information about each
transaction with the buyers and the sellers: The architecture of the system
will be such that code and data objects to be executed are passed directly
between buyer and seller, so that the arbitrators are not a bottleneck in the
transaction. The arbitrator can require
a minimum fee per trade to guarantee that its transaction execution costs are
covered, and perhaps a small fraction of the transaction value above that. We further anticipate that arbitrators are
themselves deployed in a distributed arrangement (similar to Akamai’s) to
provide fast access to the market from anywhere in the Internet’s network
topology.
It should not be difficult for OCEAN to grow
fast enough to be profitable within 5 years if a talented and well-connected
team pursues the above-described fast-growth strategy. More detailed projections are forthcoming.
The primary influence on profit margins will be
the potential for competing markets obeying the standards to pop up and offer
slightly lower transaction fees, as in the market for online stock-trading
brokerages. This danger can be
alleviated by imposing only modest profit margins to begin with, and by
encouraging potential competitors to partner with OCEAN instead of competing
with it.
The largest danger to profitability is perhaps
that a guerilla open-software P2P implementation of OCEAN will arise that has
no central arbitrators, and collects no transaction fees, yet provides the same
service, of a global virtual trading floor.
However, the same fate could have happened to eBay or to the
present stock and commodities markets (that is, a competing, free, totally
distributed version could have emerged), yet this has not yet happened. One possible explanation is that it is
difficult for a totally distributed network to provide the role of a trusted
third-party arbitrator who monitors trades and resolves disputes between
parties to a transaction: This is the role of the core OCEAN systems, the
arbitrators. However, it is conceivable
that in the future some clever theorist will figure out how to make a totally
trustworthy, distributed automated market that nevertheless maintains
consistent prices, for minimum economic friction. But, we suppose no venture is totally free of risk against future
technological advancements.
F.
The Team
The OCEAN project has had two lives so far: one
as a student-organized project and $50K competition entry led by myself and
assisted by fellow students from MIT and Stanford, circa 1996. Although the project generated a lot of
excitement locally, none of us had the time to pursue it beyond the
competition: I myself had to (or wanted to) finish my Ph.D. in the physics of
computing.
OCEAN's second lease on life started in the last
year: After receiving my Ph.D., as I entered my new faculty position at the
University of Florida's Computer & Information Science & Engineering
Department, I decided to restart the project.
Here are the people who are or have been
involved with OCEAN:
First, a brief bio of myself:
My students who have contributed to the project
so far include:
Several other UF graduate and undergraduate
students have expressed in interest in participating in the OCEAN project in
future semesters.
Some members of the past MIT-Stanford OCEAN team
might be recruited to contribute to the project once again. These individuals include:
Although the above present and potential team members have impressive credentials, an even greater level of experience and talent may be required in the OCEAN team in order to develop the concept to its fullest potential. I personally know a few other suitable people I can try to recruit (e.g., Henry Baker, one of the founders of Akamai).
G.
The Offering
Obviously, the above "business plan
sketch" is very preliminary, and many details remain to be fleshed out and
quantified. At this point, interested
VCs should probably respond primarily with encouragement and/or feedback.
If, however, an investor wishes to see this
project develop more rapidly, then I would consider accepting 50% support of my
academic salary for a year (roughly $75K including benefits and overhead),
support for at least two half-time graduate students (~$25K each with tuition
and overhead) and access to incubator resources, in order to focus on
developing the technology and standards, refining the business plan, and
recruiting team members. Additional
funding could be useful to attract heavy-hitting team members to help refine
the business plan. Equity levels in the
venture would need to be negotiated with the technology licensing offices of UF
and possibly also MIT.