The OASIS team from INRIA has announced the calculation of the nQueens problem for n = 25 solved with ObjectWeb ProActive (2,207,893,435,808,352 total solutions found). This is a new world computation record, solved with Java, using a grid technology.
- Posted by: Francois Letellier
- Posted on: August 03 2005 10:22 EDT
The nQueens problem consists to place n queens on a nxn chessboard with that no queens are enable to capture each others.
The computation was achieved in Peer-To-Peer mode (P2P), just using the spare CPU cycles of INRIA desktop machines. As such, the computing platform was highly heterogeneous: Linux, Windows, various JVMs, PII to Xeon bi-pro from 450 Mhz to 3.2 GHz, etc.
The total duration time was slightly over 6 months (4444h 54m 52s 854), starting October 8th until June 11th, using the spare CPU cycles of about 260 machines. The cumulative computing time was over 50 years: 53 years 2 days 16 hours 27 minutes 1 seconds 117 ms!
This is a huge milestone for Java grid computing, especially considering the general opinion that Java's raw computational capabilities aren't very good.
I remember doing this in the old days, at shcool, in PROLOG.
ProActive seems to be a nice tool.
You are saying that ProActive was using spare CPU cycles, could you point out the documentation where this is explained please?
ProActive was using spare CPU cycles, could you point out the documentation where this is explained please?
Actually, we achieve this by running only the night and week ends, except for some 24/24 machines like those from the team. And we run at "nice 19".
You can find more information about the P2P infrastructure here: http://www-sop.inria.fr/oasis/ProActive/doc/api/org/objectweb/proactive/doc-files/p2p.html
Desktop grids and WORA executables are a perfect match. So it's odd that JavaGrande.org is dead and hasn't been updated in over two years. Java numerics is stillborn, and INRIA's choice of a symbolic problem confirms it.
It's merely coincidence that ObjectWeb was the platform. Condor has had what it calls a "Java universe" for years. Most of the commericial grid kits embraced Java years ago, and the Global Grid Forum's OGSA reference implementation has relied mostly on Java for a couple years. I've never seen a grid that didn't have at least two versions of Java installed.
In the Spring I interviewed at MassivelyParallel.com. As usual I didn't get the job, but the founder confirmed that the future supercomputing is farms of cheap PCs running Linux on cheap LANs, with his personal strategy being a patented software algorithm for minimizing ethernet switch contention. It's refreshing to see someone betting against high performance network gear and winning federal funding. Their showcase application was a BLAST portal which isn't innovative, but I dig that bioinformatics gets all the glory. A have to wonder about a shop betting wholly on C coding, though. Bright folks with such an intimate familiarity of Moore's and Amdahl's Laws should be clamoring for virtual machinery. Gene Amdahl sits on the board.
If Java becomes the lingua franca of utility computing, how can Sun profit?
I found your post a little cryptic and I am not sure what you mean by "ObjectWeb coincidence". ObjectWeb vision is to work on next generation of middleware at large and a framework for Grid/peer-to-peer applications like ProActive is definitively a key technology for the consortium. The ProActive team is very much involved in the ObjectWeb consortium and I am sure innovation it delivers will percolate in many other projects such as OSCAR or JOnAS in the future.