SZTAKI Desktop Grid
|This article needs additional citations for verification. (April 2016) (Learn how and when to remove this template message)|
SZTAKI Desktop Grid was initiated in early 2005 and had its public launch on 26 May 2005. The aim of the initiative was to help the Public Resource Computing model to spread via the BOINC platform in Hungary. In 2005 the level of participation in BOINC among volunteers in Hungary was very low, but the main problem was that there wasn’t any Hungary-based, open project that would use volunteer computing. The Laboratory of Parallel and Distributed Computing of the Institute created SzDG as the first public Hungarian desktop grid. It is open to host distributed research applications for any research institute in Hungary or around the world.
Local SZTAKI Desktop Grid
During the setup of SzDG the staff of the laboratory identified the security problems with BOINC that render the idea of Public Research Computing undesirable among enterprise communities. To overcome these security issues (e.g. companies unwilling to let any information be sent out to public domains) the laboratory is offering a freely available local package of the BOINC software. Wrapped in a Debian package, it only requires a single computer with the Debian operating system to create a desktop grid server in a local environment.
SzDG is an online architecture, run by the Laboratory of Parallel and Distributed Systems. The staff of the laboratory maintains the system, which is open for any scientific research (see the section on DC-API to get an idea of the characteristics of suitable applications) seeking immense computing power. SzDG currently hosts one mathematical project.
The program aims at finding many generalized binary number systems. An extensive search is performed in the finite set of matrices of given size fulfilling some necessary conditions. The difficulty is that the size of this finite set is an exponential function of the dimension. It now seems possible to attack the case of 11 × 11 matrices. To check further necessary conditions the program performs a lot of floating-point calculation. Thus, a lot of CPU time is needed. Luckily, parallelization is possible, so the project can benefit from running on multiple machines.
The program outputs a list of matrices (to be more precise, their characteristic polynomials) that are already likely to be number system bases. This list is processed by another program (which does not need so much CPU-time). The final result is then a (complete) list of binary number systems in a fixed dimension.
In beta phase, the project started by investigating the 10th dimension, which entailed the processing of ninety thousand matrices, of which a total of 383 pieces seemed to be worthy of further inspection.
The Distributed Computing API (DC-API) was created by the laboratory to help the developers of distributed applications to overcome the difficulties of program development. The API hides the idiosyncrasies of BOINC, allowing developers to focus on their own research tasks. The API comes prepacked in a Debian package, available freely from the official website indicated below.
The DC-API allows easy implementation and deployment of distributed applications on multiple grid environments.
In order to accommodate the needs of very different grid environments, the DC-API supports only a restricted master-worker programming model. The restrictions include:
- Master-worker concept: there is a designated master process running somewhere on the grid infrastructure. The master process can submit worker processes called work units.
- Every work unit is a sequential application.
- There is support for limited messaging between the master and the running work units. It can be used to send status and control messages, but it is not suitable for parallel programming.
- There can not be any direct communication between work units.
Using the Berkeley Open Infrastructure for Network Computing (BOINC) distributed computing platform.
- The current science application version is 2.06.
- At this time there is no screen saver.
- Work unit CPU times vary widely: some work units can be very fast (10 minutes) and some can be very slow (75 hours).
- Most work units will take about 45 minutes.
- Work units are around [100kb ... 135kb] each, averaging around 120kb.
- You have to run many work units to get levels of credit comparable to SETI or Climate Prediction distributed programs.