Professor Betty Salzberg elected as Fellow of IEEE
The College is pleased to announce that Professor Betty Salzberg was selected as an IEEE Fellow. See www.ieee.org/web/aboutus/fellows/new-fellows.html.
According to the citation, Professor Salzberg was elected for her contributions to the database field including "access methods, online re-organization methods, and robust application techniques in computing". Much of this work was in collaboration with Dr. David Lomet of the Microsoft Corporation and with CCIS graduate students.
The normal way business data is accessed is by creating a key for each record. The records can be stored in key order so records with similar keys are on the same disk page. However, there are special problems with spatial and temporal data. It is difficult to cluster and index spatial data in more than one dimension. Professor Salzberg's spatial indexing (called the hB-tree or Holey Brick tree) clusters data in several dimensions so that nearby data is stored on the same disk page even when there are insertions and deletions made in the collection. Her temporal indexing method (the Time-split B-tree) is an index which is optimized to answer the query: "show me the state of the data collection at a given past time." Her method limits the amount of replication while guaranteeing that a certain percentage of the records on each data page accessed satisfy the query. This work was in collaboration with Dr. Lomet.
Professor Salzberg's work on online reorganization concerns merging B-trees and consolidation of B-trees while work is ongoing. B-trees, the ubiquitous index in all major DBMS products, can cluster data by key into disk pages. But as the data collection grows, the pages cannot remain in order by key. Some pages may become sparse. Multiple insertions can cause the original ordering of the data pages to be changed. Loading new data and merging it to an existing tree can exacerbate the insertion problem. These reorganization problems have been treated in a series of Professor Salzberg's papers with some of her former students.
Most recently, Professor Salzberg has been working with Dr. Rui Wang (CCIS Ph.D. 2006) and Dr. Lomet to make an important part of database applications recoverable. Modern computer systems provide services to clients through methods run at the server. These methods can be made recoverable by logging events which are non-deterministic, such as the receipt of a message from another server. The trick is to implement this in such a way that an application programmer can simply call the service methods, without having to know anything about the recovery algorithms.