Zukunft von BOINC - Future of BOINC

[rebirther]

Hier mal ein kurzer Ausschnitt einer aktuell angelaufenen Diskussion:

Everyone makes valid points; let me put in my two cents.

For me, the big goals are to create a new paradigm for scientific computing,
to enable lots (thousands) of computational science projects,
and to involve the public in science.
How are we doing?

- The dominant paradigms (clusters, grids) have about 100,000 CPUs,
shared by about 10,000 scientists.

- Volunteer computing has about 800,000 CPUs, shared by about 50 scientists.
About 70% of the total CPU power goes to two projects,
SETI@home and Folding@home.

What's wrong with this picture??
The biggest problems are:
1) how to get more scientists to use volunteer computing
2) how to get volunteers to diversify their "portfolios"

These problems are more human than technical.
Let's concentrate on 1).
Why don't more scientists use BOINC?
I think the main reasons are:

a) They haven't heard of it; they get their computing advice
from local IT people, computer companies, etc.,
which for various reasons don't promote volunteer computing.

b) In spite of our efforts, it's very hard to use BOINC.
It requires learning lots of new technologies and writing lots of code.
The documentation is confusing.
We don't offer BOINC-enabled versions of the canned apps
that many scientists use (BLAST, MATLAB, etc.).
We don't support the languages they use, like Python and Java.
We don't have any "workflow" tools.

Has any scientist decided not to use BOINC because it lacks
P2P data distribution, or an infinitely scalable server model?
Not to my knowledge.
So we need to be very careful; if we build a technically superior
platform that nobody uses, we've failed.

Having said that, I agree with Marcin that we need to plan for the future.
This process should be driven by anticipated application requirements.
How could BOINC handle the processing of CERN's LHC data,
with its petabyte-scale data?
How could BOINC handle Google's map-reduce model?
How could a project handle 100 million clients?
And so on.

As Marcin points out, we can (and must) move BOINC into the future
while maintaining a rock-solid release
and keeping up with changes in client OSs.
I think this is possible; the BOINC architecture is malleable,
and the software is small and modular enough to handle
sweeping changes without too much pain
(we've made lots of big changes already).

So the bottom line: let's think and talk about "BOINC 2010",
but let's also talk to our scientist friends,
figure out what they REALLY need, and try to supply it.

-- David