Wednesday, February 22, 2017   22:07 EET
print version

Bioinformatics and Hardware Group

Hardware and parallel computing

In the laboratory we investigated the possibility of the implementation of some facilities of dynamic molecular simulation on configured computers for parallel work. To apply this activity implies at least two aspects:

  1. the selection and analysis of a domain for the dynamic molecular analysis
  2. the facilities offered by the varied software developments to fulfill a program on computers coupled into a cluster

In all these activities we guided our activity towards the “open source” domain with an important support of program packages used.

For the dynamic molecular simulation we use two packages of programmers GROMOS 96 (Groningen Molecular Simulation Cellular Mechanics). One of the determinant factor, which influence the performances of calculus into a system of molecular dynamic is represented by hardware-software interface. This way exists a tight relation between the algorithms designed for the parallel work, the distribution of assignments between the host processor and the processors from the cluster, processor architecture and tasks programmation. This is one of the reasons for which we analyzed the hardware availability offered on the market today concerning the parallel calculus. Recently GROMOS offers a version which may roll into a parallel system of calculus. This GROMOSP product is not yet completely optimized and offers only a part of the initial soft faculties.

The IEEE SCI standard together with interconnecting components of high level performance (e.g. Myrinet) do allow the construction calculus cluster knot with a balanced support between calculus and the performance of communication, in consequence are made suitable for the parallel processing using usual PC’s (e.g. Intel P4 uniproccesor or dual processors Pentium Pro). Among the considered types of clusters in our laboratory are the Beowulf clusters. These are cluster dialing performantions based on hardware in currently use on a classic network system with none-owner of software infrastructure (Linux).