Scalable Database Systems :  Research Projects

MultiDimensional Clustering

The goal of this project is to provide an elegant relational database
solution for clustering tables along multiple dimensions. This is very
important in the Data Warehousing and Data Mart applications since most of
the analysis tends to be dimensional. For example, a Retail data
warehouse is organized along the key dimensions of Time, Products,
Geography (Regions), Suppliers, and Customers. Analysis is performed
using some or all of these dimensions.  A whole bunch of OLAP and
Multidimensional databases sprung up in recent years targeted at this
type of analysis claiming that relational databases are not naturally
suited for such work.  With this project, we are aiming to enhance the
relational database technology so that it will perform efficiently and
without requiring complex indexes for multidimensional analysis.

Consider the TPC-D benchmark. The significant dimensions of this
benchmark are:

• Time: Dates when orders are received, committed, and shipped
• Geography: Nations and Regions of Customers and Suppliers
• Parts description: Brand, type, size of parts

 

 

Already, several benchmark implementations use the Time dimension to
Range-Partition the important tables in TPC-D.  In our project, we are
exploring the ability to cluster and partition the tables using more than one
dimension. The expected advantages are in terms of better isolation of the data
to answer the multidiemsnional queries efficiently and better manageablilty of the
database. The following diagram provides a view of how such a multidimensional
clustering of the data will help to support the complex queries of TPC-D.



In this example, we are using a 2-D clustering scheme using months of Dates and Nations.
Given a query that selects a range of dates and some nations (in a region), the processing
strategy is to select the subset of the clells that belong in the intersection of  the range of
dates and specific  nations  using compact indexes. The expectation is to minimize I/O without
incurring maintenance penalites which occur due to heavy indexing.
 

[Projects:  Multidimensional Clustering |  XML Data Access  |   Tertiary Storage  |   Database Processing  ]
 
 

 
  Back to the Scalable Database Systems page