[plans] distributed.net .plan update

Plan Man plans at nodezero.distributed.net
Mon Aug 16 20:00:03 EDT 2004


distributed .plan updates in the last 24 hours:
---

mfeiri :: 16-Aug-2004 04:11 GMT (Monday) ::

I am pleased to announce the immediate availability of festog, an all new 
cruncher to search optimal and near optimal golomb rulers.
Festog is a complete rewrite from scratch but it uses and extends (!) 
techniques learned from many previous efforts by different people and 
organizations. Festog stands for "FEiris Successor TO Garsp" which is a 
hint to its main ancestor, the currently dominant cruncher called "Garry's 
Adaptation of Rado's Search Priciples".
I started to work on this rewrite in January 2002. In September of 2002 
the first working code was previewed internally at distributed.net, 
already including a 64bit version and entirely new ways of computing 
constraints that limit the backtracking during the search for OGRs. Now, 
after more than 2.5 years in the making, I feel that it is time to share 
the results with the public.

Festog is now freely available as source code under a BSD license from 
http://www.feiri.de/ogr/

A number of lessons learned in festog have already been used to enhance 
the existing garsp cruncher in dnetc. But festog represents a whole new 
platform that can use much more powerful constraints, that make it possible 
to skip significant parts of the search space.

I have run a short sample stub 25/3-2-15-26-30-4 to show the difference on 
my 1.25 GHz PowerBook G4:
dnetc needs 48 minutes to check 54 billion nodes to finish this stub
festog needs 9 minutes to check 7 billion nodes to finish this stub

Unfortunately these powerful constraints make festog incompatible with the 
crunchers that are currently in use at distributed.net. Therefore festog 
can not be used in dnetc anytime soon. But if you are as excited about 
festog as I am, you can send me an email at nogr at feiri.de and participate 
in a quick (maybe a couple of months) search for nogr20+48 using festog.


bovine :: 16-Aug-2004 18:35 GMT (Monday) ::

As previously mentioned, the work that Kakace was doing improve the
existing OGR-P2 cores has indeed resulted in further speed
improvements for even non-PPC processors as well:

On a 64-bit AMD Opteron system:
Before = OGR-P2: Benchmark for core #0 (GARSP 5.13)
                      0.00:00:16.33 [14,123,965 nodes/sec]
After = OGR-P2: Benchmark for core #0 (GARSP 6.0-64)
                      0.00:00:16.22 [23,151,097 nodes/sec]

And even a slight gain on Intel Pentium III (Tualatin) system:
Before = OGR-P2: Benchmark for core #0 (GARSP 5.13-A)
                      0.00:00:16.60 [8,192,495 nodes/sec]
After = OGR-P2: Benchmark for core #0 (GARSP 6.0-A)
                      0.00:00:25.64 [8,709,555 nodes/sec]

There will be a number of new pre-release clients made available in
the next day or so, so eager testers can look forward to playing with
them soon! :)



More information about the plans mailing list