Using Virtual Processors and Mathematical Software for Cluster Computing Performance

Title: Using Virtual Processors and Mathematical Software for Cluster Computing Performance

Abstract: This research investigated the use of virtual processors technology and mathematical software for improving parallel programs in a cluster environment. The study used Intel Hyper Threading technology and MATLAB 6.5 Release 13 scripts for floating-points computation. The results showed that using virtual processors is a good technique for enhancing parallel programs, not only for memory-based computations but also for massive disk-storage operations. The study concluded that this approach is beneficial for improving cluster computing performances.

Main Research Question: Can using virtual processors and mathematical software improve parallel programs in a cluster computing environment?

Methodology: The study used a cluster of two nodes with Intel Xeon 3.20 GHz processors and 2 GB of RAM each. The nodes were connected via a 1 Gb switch. Each node had four SCSI disks in a RAID 5 configuration, providing 36.5 GB of storage. The study used MS Windows Server 2000 and SuSE Linux 8.1 operating systems, with MATLAB 6.5 Release 13. The research involved splitting a given computation into multiple instances of the MATLAB program, with each instance running on a node. The study used a master-slave approach, where the master instance launched the slave copies and controlled their execution. The slaves performed the same set of instructions on different sets of data. The study also investigated the exchange of messages among independent processes and the use of shared files for communication.

Results: The study found that using virtual processors and mathematical software improved parallel programs in the cluster computing environment. The results showed that this approach was not only effective for memory-based computations but also for massive disk-storage operations. The study also found that the exchange of messages among independent processes was not critical for the time execution if the nodes were connected in a fast private Lan and used fast mass-storage systems like SCSI or FiberChannel.

Implications: The study's findings suggest that using virtual processors and mathematical software can significantly improve parallel programs in a cluster computing environment. This approach can lead to faster execution times and better resource utilization. The study also highlights the importance of fast communication channels and mass-storage systems for effective parallel computing.

Link to Article: https://arxiv.org/abs/0401006v1 Authors: arXiv ID: 0401006v1