GAMMA was a highly parallel database machine developed by Prof. David Dewitt and other researchers at University of Wisconsin-Madison. While the system was initially targeted at VAX machines, it was later ported to Intel Hypercube. The hypercube consisted of 32 processors and 32 disk drives. Dewitt calls GAMMA as a successor of DIRECT.
- The storage subsystem of GAMMA was called WiSS.
- GAMMA ran on a specialized operating system called NOSE.
- Shared nothing architecture
- Each processor has its own disk and RAM
- Interconnection network enables communication between processors
- Message passing is the dominant communication strategy employed
- Horizontal partitioning(Declustering) of relations across disk drives
- Allows parallel scanning of relations
- Several relational operators can be started on each processor
- Hash based parallel algorithms are used behind relational operators
- Allows indefinite scaling.
- Dataflow scheduling techniques coordinate multioperator queries
GAMMA provided the following declustering strategies
- Round-Robin(default strategy)
- Hash based(Randomizing function on the key attribute)
- Range based(User specifies range values)
- Clustered indices in Gamma can be constructed on any attribute. VSAM and Tandem restricts such indices to partitioning attributes.
Points to Ponder
How does GAMMA compare to Google's BigTable and Map-Reduce Paradigms?