Beowulf clusters

The computational demands for Hollander s simulations (in 2000) were typically of the order of 1 CPU week on a PentiumIII/500 MHz at 400 MB of memory per run. The computer code was fully parallelized and was run on a 12 CPU Beowulf cluster. 

The MPI implementation was performed according to a method that Fischer [3] used in his incompressible Legendre hp-spectral method. The method is based on the use of three arrays that keep track of the global/local element and mortar numbering, the processor on which the local element and mortars are allocated, and the face side of the slave element of a mortar. The mortars are allocated at the same processor in which the master domain is allocated. The MPI code was tested for a laminar backward-facing step flow on the local 8-node (16-processor) Beowulf cluster and showed an efficiency of 90% for large grids (Fig. 3.1). 

Figure 3.1 Actual (i) and linear (2) trends of iterations pet hour plotted vs. the number of processors for the local 8-node dual PIII/750 MHz-processor Beowulf cluster.

To further bring down the computational cost of the simulations, the code will be parallelized. The code will run on a network of desktop computers. Each computer will perform part of the calculation and communicate its results to the others. This approach allows a collection of cheaper computers to be used instead of an expensive supercomputer. The Fluid Dynamics Group at the University of Massachusetts shares a 180 CPU Beowulf cluster for large, parallel computations. These computer resources will be able to simulate the formation of tens or hundreds of drops. Though this will only represent a fraction of the entire spray, it will certainly be sufficient to reveal the interesting physics behind spray breakup. 

The computations were done in parallel on a cheap Beowulf cluster consisting of 8 PCs with the processors AMD Athlon 1.4 GHz, 768 MB RAM. The PCs are interconnected by Fast Ethernet LAN. The results in Table 2 correspond to computations on 3 and 8 PC s from the cluster. 

Although a Beowulf cluster may not outperform the fastest parallel supercomputers in the world, the idea that you can build relatively inexpensive parallel computer environments that can be fully dedicated to your own research agenda is surely highly attractive. 

NAM and polymers A seal able molecular dynamics code that can be run on the Beowulf parallel PC cluster used to run molecular dynamics simulations on selected molecular systems... 

Memory requirement for an unstmctured CFD code such as Fluent is approximately 1 GB RAM per IM cells. Nowadays, a total RAM of the order of 5-10 GB is standard for a single off-the-shelf PC configuration, making it possible in principle to fit the entire simulation into such a machine. However, to obtain practical turnover times, a parallel computer platform would be needed to solve virtual mannequin problems. In the form of Beowulf Linux clusters, such systems are quite inexpensive nowadays and can be bought virtually off-the-shelf. The development of a reliable, full-scale virtual matmequin seems a very realistic and feasible goal for the next few years. 

Beowulf-class system A commodity cluster implemented using mass-market PCs and COTS network technology for low-cost parallel computing. 

Cluster computing. Integrates stand-alone computers devised for mainstream processing tasks through local-area (LAN) or system-area (SAN) interconnection networks and employed as a singly administered computing resource (e.g., Beowulf, NOW, Compaq SC, IBM SP-2). 

Beowulf-class systems are commodity clusters employing personal computers (PCs) or small SMPs of PCs as their nodes and using COTS LANs or SANs to provide node interconnection. A Beowulf-class cluster is hosted by an open source Unix-like operating system such as Linux. A Windows-Beowulf system runs the mass-market widely distributed Microsoft Windows operating systems instead of Unix. 

More information on how to build a Beowulf computer can be found in the book How to Build a Beowulf A Guide to the Implementation and Application of PC Clusters, featuring chapters by Sterling, Becker and many others (Sterling et ah, 1999). 

---