An Introduction to Parallel Computing

SMP, MPP, clustered SMP, NUMA, data parallelism, sharednothing and shared-everything architectures. Understanding these terms and concepts is critical to getting the best performance out of your data warehouse or data mart. It’s proven technology today to combine dozens or hundreds of CPUs to build a supercomputer to host terabyte-class data warehouses and data marts hundreds of gigabytes in size. However, largely because SAS Software prior to Version 8 existed in the “sequential” world (with the exception of SPDS), most SAS users have never had to become familiar with the parallel computing techniques that make these platforms essential in today’s business computing environments. Things are changing fast, though. With Version 8, SAS Software is capable of taking advantage of multiple processors, and third-party vendors now have software that integrates SAS into their own parallel products. This paper presents an overview of parallel computing concepts, terminology, and architectures that should provide SAS users with a working familiarity of the subject. The paper is organized into sections on parallel processing in general, then parallel hardware architectures, followed by parallel software architectures. Users working with very large databases, data warehouses, and data marts may find this paper immediately useful. Disclaimer: The views and opinions expressed here are those of the author and not those of First Union National Bank. First Union National Bank does not endorse, recommend, or promote any of the computing architectures, platforms, or products referenced in this paper.