Pipelined Hash Joins Using Network of Workstations

Drrnands for h ish transaction procrssing rates from databasc systsms Ird to the use of parallel processing techniqiiçs for transxtions and queries. Most commercial parallel database systems suffer from sxceptionally high hardware and software costs. Availùbility of public domain software (such as PVM and MPI) that hnmcsses several workstations on a LXN into a vinual pardlel machine provides a welcome alternative for parallel query processing. This thesis exploits such a Network of Workstations for intra-qurry (pipe l i n d ) and intrn-operation (partitionrd) parallelism usin_o the Hash Join algorithm in a crntralizrd qurry processing architectiire. A comprehensive performance evaluation siigsests that Pipelined Hash Joins rxhibit significant improvemrnt in response timr for complex queries and for qurries consiiming very large input relations.