To study the internals of database systems as an introduction to research and as a basis for rational performance tuning.
The study of internals will concern topics at the intersection of database system, operating system, and distributed computing research and development. Specific to databases is the support of the notion of transaction: a multi-step atomic unit of work that must appear to execute in isolation and in an all-or-nothing manner. The theory and practice of transaction processing is the problem of making this happen efficiently and reliably.
Tuning is the activity of making your database system run faster. The capable tuner must understand the internals and externals of a database system well enough to understand what could be affecting the performance of a database application. We will see that interactions between different levels of the system, e.g., index design and concurrency control, are extremely important, so will require a new optic on database management design as well as introduce new research issues. Our discussion of tuning will range from the hardware to conceptual design, touching on operating systems, transactional subcomponents, index selection, query reformulation, normalization decisions, and the comparative advantage of object-oriented database systems. This portion of the course will be heavily sprinkled with case studies from database tuning in biotech, telecommunications, and finance. Also, since the book that Philippe Bonnet and I have written has many tests associated with it, you will get the benefit of those tests.
Because of my recent research (and product) interests, this year will include frequent discussions of
We will also discuss, if we have time several related topics found at the end of these lecture notes (i) high performance replicated database systems without the use of two phase commit, (ii) case studies from Wall Street, (iii) a self-tuning technique that improves on LRU, and (iv) a data structure for data warehouses.
Here are some experiments having to do with database tuning .
Here find a brief lecture on the paper "Easy Impossibility Proofs for Distributed Consensus Problems" by Fischer, M. and Lynch, N. and Merritt, M. Here is a relevant picture.
Here is how to call C from K: Don Orth's description of how to call C from K.
Here are Alan Fekete's slides on snapshot isolation Snapshot Isolation and Fixes to It
Here is Joe Conron's nice paper on indexes (from when he was a master's student).
Some results from database tuning projects. Presented as rules of thumb. Here is one very nice tuning project by Yuhong Chen. Here is another by Ilya Finkelshteyn Here is a third by Marina Balina
Here is an approximation to an n-server capacity planner.
Here are Alberto Lerner's excellent notes on performance monitoring. Here you can find his thesis.
Here are notes about materialized views in Oracle.
Here is a call to a new organization of databases by the Turing Award winner Jim Gray
Here are two very nice and very practical papers about data warehousing by one of the foremost practitioners in the field: Ralph Kimball on constructing an enterprise data warehouse and Ralph Kimball on why to use star schemas in dimensional modeling.
Finally, here are the rules about academic honesty.