Page 593 - (ISC)² CISSP Certified Information Systems Security Professional Official Study Guide
P. 593

Large-Scale Parallel Data Systems


               Parallel data systems or parallel computing is a computation system
               designed to perform numerous calculations simultaneously. But
               parallel data systems often go far beyond basic multiprocessing
               capabilities. They often include the concept of dividing up a large task
               into smaller elements, and then distributing each subelement to a

               different processing subsystem for parallel computation. This
               implementation is based on the idea that some problems can be solved
               efficiently if broken into smaller tasks that can be worked on
               concurrently. Parallel data processing can be accomplished by using
               distinct CPUs or multicore CPUs, using virtual systems, or any
               combination of these. Large-scale parallel data systems must also be
               concerned with performance, power consumption, and

               reliability/stability issues.

               Within the arena of multiprocessing or parallel processing there are
               several divisions. The first division is between asymmetric
               multiprocessing (AMP) and symmetric multiprocessing (SMP). In
               AMP, the processors are often operating independently of each other.
               Usually each processor has its own OS and/or task instruction set.
               Under AMP, processors can be configured to execute only specific

               code or operate on specific tasks (or specific code or tasks is allowed to
               run only on specific processors; this might be called affinity in some
               circumstances). In SMP, the processors each share a common OS and
               memory. The collection of processors also works collectively on a
               single task, code, or project. A variation of AMP is massive parallel
               processing (MPP), where numerous SMP systems are linked together

               in order to work on a single primary task across multiple processes in
               multiple linked systems. An MPP traditionally involved multiple
               chassis, but modern MPPs are commonly implemented onto the same
               chip.

               The arena of large-scale parallel data systems is still evolving. It is
               likely that many management issues are yet to be discovered and

               solutions to known issues are still being sought. Large-scale parallel
               data management is likely a key tool in managing big data and will
               often involve cloud computing, grid computing, or peer-to-peer
               computing solutions. These three concepts are covered in the
   588   589   590   591   592   593   594   595   596   597   598