Two DIKU Talks: DBToaster and and Languages for Mathematical Computing – Københavns Universitet

Two DIKU Talks: DBToaster and and Languages for Mathematical Computing

DIKU Talk 1: Lille Aud, Sep 7, 2012, 13.00-14.00


DBToaster: Higher-Order Delta Processing for Dynamic, Frequently Fresh Views

by Yanif Ahmad, Johns Hopkins University


Many of today's popular computing applications require online and real-time analytics over large and dynamic datasets, from social web applications, to electronic markets, enterprise auditing and scientific simulation and analysis.  DBToaster is a query compiler enabling applications to handle the combination of large data and rapidly changing data by transforming declarative database views and analysis queries into efficient incremental stream processing code.

In this talk I will present the recursive finite differencing technique at the core of DBToaster called the viewlet transform. Our viewlet transform materializes both user-specifed queries for incremental maintenance, as well as a supporting set of higher-order delta views, for a substantially lower view maintenance cost. These higher-order delta views are amenable to implementation with compilation to machine code. In many cases, compilation can avoid using classical query operators such as joins and eliminates the prevalent interpreter overheads in database engines. DBToaster achieves tens of thousands of complete view refreshes a second for a wide range of queries, providing 3 orders of magnitude speedup over alternative state-of-the-art systems and enabling new dynamic data processing applications.

I will also briefly discuss ongoing work in developing an intermediate trigger processing language alongside DBToaster, called K3, for building highly flexible data processing runtimes and direct embedding of database technologies in user applications, as well parallel execution of our trigger programs for extremely large distributed views.

DBToaster ( has been jointly developed at Johns Hopkins University and EPFL with Oliver Kennedy (soon to be at U. Buffalo) and Christoph Koch.


Yanif Ahmad is an Assistant Professor in the Computer Science Department at the Johns Hopkins University. He previously held a position as a postdoctoral associate in the Database Group at Cornell University, and received his Ph.D. with the Database Group at Brown University in 2009.

Yanif's research interests span stream processing engines, declarative languages and large-scale database systems, and seeks to bridge databases, distributed systems, and programming languages. His current projects focus on incremental systems, declarative query optimization and model-based databases. Yanif is the recipient of several awards, including an IBM Ph.D. Fellowship, ICDE 2008 Best Paper Award, and
SIGMOD 2005 Best Demonstration.

DIKU Talk 2: Small Auditorium, Sep 7, 2012, 14.00-15.00

A Cluster of Languages for Mathematical Computing

by Stephen M. Watt, Western University, Canada


The modern computing landscape features a galaxy of programming and data description languages, with new ones introduced continually.  New ideas, introduced in one language, find their way into subsequent languages as the landscape evolves.  Many important ideas first expressed in mathematical software have found their way into the most widely used general purpose languages.  Perhaps one reason for this is
that mathematical programming provides a rich domain of challenging, but precisely defined problems.  This contrasts with the hard problems in other areas where abstractions are simpler and it is harder to evaluate different approaches.  We therefore see mathematical software as a canary in the coal mine of programming languages, providing an advance testing ground for ideas.

We present our experiences in the design and implementation of several special purpose languages applied to mathematical computing: Maple, Axiom, Aldor, OpenMath, MathML and InkML. These languages have addressed a breadth of problems, ranging from efficient compilation of higher order mathematical abstractions to flexible motion capture of two dimensional handwriting. In terms of adoption, they range from tens to millions of users with widely differing needs. This talk outlines some of the problems these languages were intended to solve, the new ideas they introduced, and the pragmatic compromises taken along the way.


Stephen Watt is Distinguished University Professor at Western University, Canada.  His research lies at the cross-roads of mathematics and computing, spanning both its theoretical and practical aspects.  He is one of the original authors of both the Maple and Axiom systems for computer algebra and principal architect of the Aldor programming language and its optimizing compiler.  He has co-authored the MathML and InkML W3C Standards and helped open the area of symbolic-numeric algorithms for polynomials, a current hot topic in computer algebra.  

More recently, he has focused on open-based computing and algorithms for polynomials of symbolic degree. He has been active in a number of early stage software companies and is presently lead independent director of the Descartes Systems Group, a publicly traded company offering SaaS solutions for logistics and
global trade.  He has received numerous awards for his research work and his work with companies.

Both talks are open to all interested people and free of charge.

Water will be available, but no further refreshments will be served until later, since the talks will be followed by a PhD defense where most of the audience will be present.