PARALLEL PROCESSING WITH CLUSTERED WORKSTATIONS

NOWADAYS, MOST ORGANIZATIONS USE WORKSTATIONS THAT ARE ATTACHED TO A NETWORK TO CONDUCT THEIR DAILY OPERATIONS. THE USE OF A CLUSTER OF THESE WORKSTATIONS AS A PARALLEL PROCESSING SYSTEM CAN PROVIDE A VERY POWERFUL COMPUTATIONAL CAPABILITY AT A VERY LOW COST. HOWEVER, THE MAJOR IMPEDIMENT TO THE USE OF EXISTING TECHNIQUES FOR CLUSTERING WORKSTATIONS INTO A PARALLEL PROCESSOR IS THE EXCESSIVE PERFORMANCE OVERHEAD INCURRED BY THESE TECHNIQUES. EXISTING TECHNIQUES DELIVER ONLY A SMALL FRACTION OF THE POTENTIAL COMPUTATIONAL POWER OF A CLUSTER OF WORKSTATIONS. RESEARCHERS ARE ADDRESSING A NEW APPROACH FOR CLUSTERING WORKSTATIONS CALLED THE NETWORK SHARED MEMORY APPROACH THAT IS ATTRACTIVE BECAUSE OF THREE REASONS: (1) IT IS CAPABLE OF DELIVERING PERFORMANCE THAT IS AT LEAST AN ORDER OF MAGNITUDE BETTER THAN THAT ATTAINABLE USING EXISTING TECHNIQUES FOR CLUSTERING WORKSTATIONS; (2) IT GREATLY SIMPLIFIES THE PROGRAMMING TASK BY PROVIDING A SHARED MEMORY ABSTRACTION TO THE PROGRAMMER; AND (3) THE COST OF ATTAINING THIS PERFORMANCE AND PROGRAMMABILITY IS SEVERAL TIMES LOWER THAN THAT OF COMMERCIALLY AVAILABLE SHARED MEMORY PARALLEL PROCESSORS OF COMPARABLE COMPUTING POWER.