[Announcements | Schedule and Readings | Assignments and Quizzes | Syllabus/Policies ]
Reading Questions, Lecture 8 (10/4)
Papers: Parallel Database Systems,
Optimizer Validation and Performance Evaluation in R*
These papers deal with issues in multiprocessor database systems. The Dewitt and Gray paper is a high level summary of database architectures for parallelism, illustrating some of the techniques that can be used to exploit the availability of multiple processors. The R* system (not to be confused with R* trees!) is an extension of System R to run on multiple machines -- they are particularly concerned with the additional cost of network I/O in query processing.
Questions to consider:
- What's the difference between a parallel and a distributed database? What issues are different in
one architecture versus the other? In what ways are the two architectures alike?
- Why do Dewitt and Gray advocate a shared nothing architecture?
- In what ways must existing database architectures be modified to support multi-processor environments? What new data layout issues are introduced? What new query processing challenges must be addressed?
- In System R*, which distributed join algorithm performs the best? Why? Do you think it would still perform better than the others in a modern database system?
Samuel Madden (madden at csail dot mit dot edu)
Last modified: Thu Sep 30 18:34:01 EDT 2004