Although we belong to very different cultural backgrounds there is something that unites Computer Scientists: knowledge and the communication of new ideas and thoughts! I hope you enjoy reading my Web Blog. Please feel free to leave comments and second thoughts. This time I am including one of my latest reviews at ACM Computing Reviews, a review about a very complex paper:

Shao, F., Novak, A., and Shanmugasundaram, J. 2006. Triggers over nested views of relational data. ACM Trans. Database Syst. 31, 3 (Sep. 2006), 921-967. DOI= http://doi.acm.org/10.1145/1166074.1166080

Available at: http://portal.acm.org/citation.cfm?id=1166080&dl=acm&coll=&CFID=15151515&CFTOKEN=6184618#

This article is all about triggers and nothing but a special kind of triggers: triggers over nested virtual (non-materialized) xml views that represent nested relational views on top of object-relational DBMS’s. According to both the SQL99 and SQL2003 specifications, triggers as schema objects, are only specified on base tables. Therefore both specifications do not contemplate the existence of triggers on views. More over the SQL2003 ISO/IEC JTC 1/SC 32 draft standard on XML-Related Specifications SQL/XML does not include a single reference to trigger inclusion.

We might have to wait till the ISO’s Work Group for database languages (WG3) schedules in 2007 the development of the next ISO SQL 2007 standard which might include triggers on top of views as presented in this paper in the framework and foundations parts of the specification and in the XML-Related Specifications SQL/XML part, the inclusion of triggers in xml views.

In summary the paper describes the entire round-trip of a middleware active triggering system of virtual xml views expressed on top of object-relational base tables. The round trip can be described as follows:

1. An update occurs on base tables

2. The system detects how updates on base tables affect triggers expressed on top of virtual xml views

3. Activated triggers expressed on top of nested virtual xml views representing nested-relational unmaterialized views are translated into plain SQL triggers on top of object-relational base tables

4. Base table triggers are executed and results are re-converted into nested relations passing as parameters to the affected virtual XML View triggers.

Several important contributions are indicated by the authors in this large 47 pages journal article, among them we may cite the following:

1. A systematic technique for the translation of triggers on top of NON-1NF nested-relational views into SQL triggers of flat tables

2. An algorithm for detecting changes in nested views

3. The definition and use of a special type of views entitled as “injective views” capable of detecting in an optimized way old values and new values in triggers with regard to transition tables and trigger execution context.

4. The use and adaptation of trigger grouping optimization techniques used in publish-subscribe systems for the definition of triggers on top of virtual xml views.

In their experimental evaluation with syntethic data sets we conclude that their techniques do not scale well for materialized views and for non-grouping techniques whereas using the adapted trigger grouping techniques in virtual views scales efficiently in standard commodity harware (an already outdated 933Mhz PIII 1GB RAM Computer).

I’d rather say that the work of the authors have been presented in several VLDB conferences and that their contribution to the field is important which might be taken into account for the definition of the next ISO SQL 2007 standards, especially with regard to views. According to some slides from Jim Melton, the next SQL 2007 standard will discuss the inclusion of materalized views but apparently there is no prior agenda containing virtual xml views, triggers on top of relational views and triggers on top of xml views.

I only have to confess one negative criticism to the authors work with regard to the definition of a continuum of examples to prove the versality and universality of their techniques and more importantly decidability of trigger execution, which is a non-trivial very complex problem.