parallel_computing[1]

parallel_computing[1] - Parallelization lets applications...

Info iconThis preview shows pages 1–3. Sign up to view the full content.

View Full Document Right Arrow Icon
COMMUNICATIONS OF THE ACM September 2007/Vol. 50, No. 9 75 “The first law of massive parallelism is the foundation for massive marketing that sup- ports massive budgets that supports the search for massive parallelism,” Gordon Bell, 1992 [2]. For many years parallel computers have been used by an exclusive scientific niche. Only rich universities and research institu- tions backed by government budgets or by multibillion-dollar corporations could afford state-of-the-art parallel machines. Multiprocessor machines are very expensive and demand highly specialized expertise in systems administration and programming skills. Commodity workstations first appeared in the 1990s. New computer networking technologies allowed the harnessing of tens, and later hundreds, of them together to form clusters of workstations. The “do-it- yourself” Beowulf clusters represented great progress (www.beowulf.org), and many more colleges and universities established parallel computing labs. Beowulf clusters are attractive because they are cost-effective, easy to construct, and scalable. They are built from relatively inexpensive, widely available components; most systems admin- istrators have the skills necessary to install and support clusters. If the processing power requirement increases, the performance and size of a Beowulf cluster is easily scaled up by adding more computer nodes. Beowulf clus- ters represent the fastest-growing choice for building clusters for high-performance com- puting and networking. As of September 2006, 361 systems (72%) were categorized as clusters in the list of TOP 500 supercom- puters (www.top500.org). Unfortunately, this achievement did not PARALLEL COMPUTING ON ANY DESKTOP ± By Ami Marowka illustration by Robert Saunders Parallelization lets applications exploit the high throughput of new multicore processors, and the OpenMP parallel programming model helps developers create multithreaded applications.
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
76 September 2007/Vol. 50, No. 9 COMMUNICATIONS OF THE ACM increase the adoption of parallel computing. Beowulf clusters and supercomputers assembled from off-the-shelf commodity proces- sors are still expensive, compli- cated to manage, difficult to program, and require specialized knowledge and skills. The batch processing involved preserves mainframe methods rather than making them more interactive and user friendly. The computing industry has been ready for the parallel com- puting era for more than a decade. Most small-to-mid-size organiza- tions use multiprocessor servers; commercial databases (such as Oracle and Microsoft SQL servers) support parallelism; Linux and Microsoft Windows Server System operating systems are multithreaded; and programming languages (such as Java) support multithreaded programming. How- ever, the massive breakthrough of parallel computing many have been waiting for has still not occurred. Two things were missing until 2005: low-cost parallel
Background image of page 2
Image of page 3
This is the end of the preview. Sign up to access the rest of the document.

Page1 / 4

parallel_computing[1] - Parallelization lets applications...

This preview shows document pages 1 - 3. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online