The MSc exam of Ella Rabinovich

Yesterday I chaired the MSc exam of Ella Rabinovich, who finished her thesis in joint supervision of Avi Gal (who was my PhD student) and myself.  Looking at my own record, Ella is my 21st MSc graduate, in addition to 6 PhD graduates (one person, David Botzer, has done both MSc and PhD theses under my supervision - so the list comes up to 26 people).    I am doing graduate students supervision as a hobby (the Technion does not pay for it),  it enables to investigate idea that are typically one or two steps ahead from what we are doing in the IBM Haifa Research Lab (which by itself has to be some steps ahead of the markets), Ella works also in our Proton team in IBM,  the thesis deals with pattern rewriting as one of the means of performance improvement in event processing run-time,  I have already written about our DEBS'11 paper on this topic.   the thesis was well accepted by the examination team and got a relative high grade.   My guess is that we'll hear more about notable  research contributions of Ella in the future. 

On pattern rewriting - DEBS 2011 presentation

Among other things I have done in DEBS 2011, I have also delivered one talk to present the paper on patterns rewriting, the paper, like all DEBS 2011 papers can be obtained from the ACM Digital Library. 

The paper is co-authored by Ella Rabinovich (the primary author), Avi Gal and myself.

The presentation starts by classifying optimization tools in use for optimizing the performance of event processing applications, the classification is into blackbox optimization where the actual implementation is taken as a blackbox, and whitebox optimization,      The paper deals with one of the types of whitebox optimization - pattern rewriting, which mean rewrite a pattern into a collection of other patterns that yield equivalent results: same output to the same input. 

 While rewriting exist in different areas such as rules or SQL queries,  there is some inherent complexity in some of the event processing patterns such as SEQUENCE.    The difficulty stems from the temporal relationship imposed by sequence, which imposes restriction on assertion splitting, and from the fact that policies may not be transferred in the rewriting process nicely.   This was an exploratory research to see whether rewriting is a useful method for optimization, especially where complex sequences are involved.  The experimental results in simulation for sequence with 16 operands (a real application we have done in the past in the web commerce area),  shows that rewriting alone can improve the latency by more than tenfold and also the throughput, and no matter how the latency/throughput trade-off is weighted,  some rewriting will give better result than no rewriting in this case.   

The presentation can be viewed on slideshare.  Enjoy!