Abstract

Today personal computers become more powerful and are distributed worldwide, getting more attention as performers of distributed computations. Connected into a local network, they act as a powerful and cheap computing resource - so called heterogeneous cluster.

There are several specifications, that implement different approaches for distributed computations, starting from simple parallel programming ending with developing a complex distributed computing systems.

It's obviously, that the fee for high abstract level and genericity of object-oriented philosophy is a performance loss. After all, high abstract level of distributed computation doesn't allow full control on data exchange operations and often demands supplementary data conversion. Therefore, before migration to high abstraction level it's reasonable to know the exact performance loss relative to the low one. It is also good to know the scope of the tasks that are expedient for such migration.

The master thesis objective is researching capabilities of implementation the given computation task on the target cluster.

The given computation task lie in simulation of growing population of biological cells in modeling environment Diana. The main characteristics of such simulation consist in mass computation of same type objects, which might have some connections between each other. The computation of a single cell state requires solving of ca. 40 differential-algebraic equations. The cell state can change step-wise depending on parameter set. At that time the equation set change of that cell occurs. One of cell states causes a division of the cell. Thus, the system is growing.

The target cluster consists of PC set, that are connected to local area network of an Institute. Since PCs are workstations of institute employees, there occur some problems that relate to insufficient stability of the computing network. PC can suddenly become disconnected from the computation process (for instance because of rebooting the system)

The problem like this splits up into two sub problems:

1. To create media for simulating a single cell

2. To research and to develop media, that will manage a group of the cells

For mathematical description of the cell, it was decided to create individual type of a model, so called "hybrid" model. The "hybridity" of the model means, that model is able to change its own equation set at certain time when certain event occurs.

Simulation of a group of cell models (so called aggregate model with varied structure) is performed with distributed solver. The distributed solver should also manage the number of hybrid models and provide stability of simulation process on several PCs.

At present time (June 2007) simulation tool Diana doesn't have any functioning solver. Therefore, this workings on hybrid solver and distribute solver have great practical value.