News archive

December 24, 2015, 14:30 MSD
New application aimed at testing stability of the GULP code.GULP is a program for performing a variety of types of simulation on materials using boundary conditions of 0-D (molecules and clusters), 1-D (polymers), 2-D (surfaces, slabs and grain boundaries), or 3D (periodic solids). Adaptation of this code for BOINC is a first step to implementation of the computational materials discovery methods to the distributed computing. More specifically, USPEX evolutionary algorithm. This is an algorithm for crystal structure prediction, which is a problem of global optimization, i.e., finding the global minimum of the free energy of the crystal (per mole) with respect to variations of the structure. Successful solution of this problem requires local optimization for each structure generated during prediction. GULP is one of the methods supported by USPEX. That is why we have adapted it for BOINC and providing a test caclucation for it. This is variable cell calculation of the crystal structure of magnesium aluminate (MgAl2O4) with 28 atoms per unit cell at 100 GPa pressure.

October 03, 2014, 20:00 MSD
We submitted a new series of workunits (12289 wus). Each Wu runs local search for 32 times. So in total 131072 local searches are expected. The precision was set to 0.00001. 10 free parameters for Tersoff potential are fitted.

October 03, 2014, 20:00 MSD
We submitted a series of workunits (4096 wus). Each Wu runs local search for 32 times. So in total 131072 local searches are expected. The precision was set to 0.00001. 10 free parameters for Tersoff potential are fitted.

October 03, 2014, 20:00 MSD
OPTIMA@home now hosts new application aimed at fitting parameters of the Tersoff potential. Molecular dynamics (MD) is often used for modeling various properties of crystals. The essence of this method consists in the numerical solution of the differential equations describing motion of atoms in the crystal lattice to determine the steady state. For molecular-dynamic computations, you must know the potential - a function that determines the interaction energy of atoms in the lattice. Modern potentials used in the simulations contain several parameters. Specific parameter values determine what concrete material is modeled. For example, for silicon, this will be one set, for diamond another etc. The project is aimed at fitting parameter values of the potential in order to achieve known material properties (energy of the lattice cell, the components of the elastic modulus etc.). Then this potential will be used for molecular dynamics simulation. To identify potential parameters we solve the problem of minimizing the variance of the values of the properties obtained by using the potential from known values. This task is to generate a large sequence of initial approximations, from which further methods local search is a (local) minimum. Next, from the obtained values we select tuples with objective values less than the specified precision.

July 08, 2014, 20:00 MSD
It is need for the optimization of the solver smallexpx.

July 18, 2013, 00:01 MSD
Since March, the project carried out calculations associated with the application smallexp. The scientific problem of smallexp is the experimental calculation of the asymptotic growth of the maximum possible number of fragments in a sequence fixed length. It is important to get information on possible periodic structures. This is necessary for the development of efficient algorithms search for fragments of small order in character sequences (the order - the ratio of the length of the words to the minimum period). It's important for discrete mathematics. Smallexpx application has more efficient algorithm then smallexp application.

Articles about scientific problem:

R. Kolpakov, G. Kucherov, On Maximal Repetitions in Words, J. Discrete Algorithms 1(1) (2000), 159–186.

R. Kolpakov, G. Kucherov, Periodic structures in words, chapter for the 3rd Lothaire volume Applied Combinatorics on Words, Cambridge University Press, 2005.

M. Crochemore, C. Iliopoulos, M. Kubica, J. Radoszewski, W. Rytter, T. Walen, Extracting powers and periods in a string from its runs structure, Lecture Notes in Comput. Sci. 6393 (2010), 258–269.

January 19, 2012, 21:00 MSD
New algorithm brought new results. New minimum is much better vs previous: -685.809. New algorithm uses the results of the previous computations for generating new work units. And it worked!

January 9, 2012, 12:00 MSD
The large experiment was finished and the previously found minimum improved to the value of: -683.84.

October 7, 2011, 22:01 MSD
New results of the recent experiments are obtained: the newly found minimum is -683.783 which is significantly better w.r.t. the previous one (-681.871). The experiment started with a set of 4096 random points. The obtained improved points served as starting points for further processing. The process finished when no further improvement is possible. This experiment lasted for one month.

August 26, 2011, 22:01 MSD
New work is submitted. We improved the algorithm: the initial portion is 4096 WUs and new WUs will be added during the computation based on the computed results. Thus we'll keep the system busy all the time until we decide to stop the computation.

August 25, 2011, 17:31 MSD
We obtained new minimum value: -681.871. Thanks all crunchers! We are working on improving the distributed algorithm. The next portion is expected the beginning of the next week.

August 14, 2011, 23:30 MSD
The next portion of WUs is submitted. This is the 3rd generation of points: we are trying to improve starting from the results obtained at the previous step. Now we also increased the number of steps in the Monotonic Basin Hopping Algorithm from 2048 to 4096 and adjusted WU settings respectively. We also set the maximum number of WUs per core in progress to 8. This is needed to improve the scheduling. The deadline is set to 3 days. All these things are done in response to the crunchers requests.

August 14, 2011, 23:10 MSD
The job submitted on Aug, 10 just finished - thanks everybody! New job is coming soon. The results are optimistic: we go -680.222 optimum value. This is again better than the previous value (-678.083). This result is due to the improved randomization policy.

August 10, 2011, 4:15 GMT
We improved the search algorithm and prepared new (2.0) application versions. The MacOSX version now supports 10.4 system. We re-run the previous experiment in order to see the performance of the improved search algorithm.

August 10, 2011, 3:10 GMT
As expected the new series of experiments brought new results: now the best minimum is -678.083 - considerably better than it was before. We are now preparing new version of application and are going to put it to the project soon as well as new workunits.

August 3, 2011, 13:10 GMT
The version for Mac OSX platform was prepared tested and deloyed.

August 3, 2011, 13:05 GMT
The new series of work units constructed from the results of the previous experiment submitted. We are expecting a considerable improvement.

August 3, 2011, 12:00 GMT
The experiment launched July, 21 has finished! The objective value was improved again and now it is -677.31. It's a remarkable improvement since the last time. Now we are working on a new experiment that will base on the results of this experiment.

July 21, 2011, 12:05 GMT
Dear friends! I've just submitted the new experiment of 16384 WUs. Moreover each WU is now 4 times "heavier" that it was before. So lets see how this experiment will pass. We are attacking 150-atom Morse cluster again...

July 21, 2011, 10:05 GMT
Experiment #2 successfully finished! Thanks everybody! After processing twice more starting points the objective value was improved: -673.769. This time the infrastructure performed much better because of more WUs and more computers registered. New experiment is coming soon! However again the last day the performance was quite low because no unprocessed workunits remained. It looks like twice more WUs or even four time more WUs are needed to load the system adequatly.

July 15, 2011, 10:05 GMT
Mac OSX version is temporaly disabled. It was incompatible with the input data and caused numerous errors. We'll build new version and enable it again soon.

July 14, 2011, 23:35 GMT
New experiment has started. Now we submitted 8192 workunits for the same 150-atom problem. I hope this number of WUs will make the infrastructure to work more efficient! The result will be published after the experiment finishes.

July 14, 2011, 23:35 GMT
Hello, friends! The first meaningful experiment finished!!! It was and attempt to locate energy minima for Morse 150-atom cluster. We submitted 4096 work units. Each WU performed iterations of Basin-Hopping method from a randomly generated starting point. The maximal_no_improve parameter value was set to 1024. The found minimum was -670.768. This experiment was quite long: it took about 1 week to complete all workunits. And most of the time system worked with only little load (see yellow line at the diagram). This very long tail is caused by results those are kept by the client machines for a very long period. The last two days only 16 results were computed! In the next series of experiments we are going to fix this by providing more workunits and stopping computations as soon as performance reaches its maximal value.

July 8, 2011, 12:35 GMT
New experiment with 4096 WUs started. The goal is to find the optimal conformation of 150 atom Morse cluster. The estimated run time was set to approximately 2 hours and the deadline increased to 2 days in order to increase the project performance: this change was suggested by zombie67 and AMDave.

July 1, 2011, 12:35 GMT
Statistics is now exported (important for credit accounting). Thanks to Phil and other users who requested this feature and provided a kind support for me.

July 1, 2011, 12:35 GMT
Message Boards are now enabled. Thanks to Phil and other users who requested this feature.

June 16, 2011, 10:30 GMT
The project will go down for maintainance from today, 12 AM to Monday, June 20, 10 AM (Moscow time)

June 13, 2011, 10:00 GMT
3rd DEGISCO (Desktop Grids for International Scientific Collaboration) project meeting is starting at London

June 10, 2011, 13:00 GMT
Statiscs charts that help you to see the current project performance are now enabled

June 06, 2011, 11:00 GMT
Do not forget to register for the Distributed Computer Infrastructure Summerschool in July in Budapest. It featurs a complete EDGI technology day. But you learn also about DEGISCO Desktop Grids, EMI middleware and StratusLab Clouds.

April 12, 2011, 11:50 GMT
The DEGISCO ISA RAS team - founding member of IDGF - participated in a PAVT'2011 which is commonly recognized as a largest HPC event in Russia. The main objective of the conference - provide an opportunity to discuss the future development of parallel computing technology and the results obtained by leading research groups in HPC. Conference topics cover all aspects of high performance computing in science and technology, including applications, hardware and software, specialized languages and packages. The talk Increasing computing power of distributed systems by grids of PCs given by ISA RAS received much attention and resulted in an interesting and fruitful discussion. The usual ISA RAS tutorial on desktop grids collected five attendees which gave high marks to the tutors and appreciated the training.

March 1, 2004, 13:50 GMT
Sample news item

Home | My Account | Message Boards

Copyright © 2019 Institute for Systems Analysis of Russian Academy of Sciences