Sprungmarken

Servicenavigation

Hauptnavigation

Sie sind hier:

Hauptinhalt

Real-Time Pattern Matching with FPGAs — VLDB 2010 Reviews

Reviews for paper Click Stream Pattern Matching with FPGAs, submitted to VLDB 2010.

Overall rating: reject

Reviewer 1

Overall Recommendation

Accept

Detailed Comments

The paper focuses on monitoring click-data streams for pattern matching. The paper is well-written -- the presentation is good. The developed system is impressive. The authors give some good examples, which are helpful to understand the system concept.

Reviewer 2

Overall Recommendation

Neutral

Detailed Comments

Summary:
---------
The demo talks about complex event detection systems, which is a topic that was largely discussed in the literature. However, the new thing in this paper is that it proposes doing this functionality in the Field programmable gate arrays (FPGA), which will result in better performance and lower response time which is major factor in complex event processing systems.

Comments:
------------
A web server can easily be used to pattern matching. The need to use a hardware is not well presented in the paper.
The demo tracks the click stream of each user separately, the authors did not specify the number of users that can be supported using one chip nor how to accommodate a larger number of the limit.
The problem of using FPGA is that every time you have a pattern, you need to create the VHDL code for it and compile, synthesize it, do placing and routing which are time and money consuming processes, and may not be practical for the click stream application.

Reviewer 3

Overall Recommendation

Neutral

Detailed Comments

This paper proposes to perform complex event detection queries, in particular, click stream patterns on FPGAs. It shows how an FPGA can be used to monitor web server data streams from the network and detect click stream patterns in real-time.
The use of FPGAs to accelerate the processing of queries and improve parallelism is an interesting idea and a few recent works have been proposed. However, FPGAs require building custom-made hårdware (eg. the query compiler) which is not trivial. Since specialized components need to be built, I am not sure how general the proposed approach is. The authors should include some discussion about the benefits but also the limitations of such an approach.
Although FPGAs have the obvious performance improvement over commodity systems, what would be the benefit of FPGAs over more expensive systems such as multicores?
It seems to me that the usefulness of the FPGAs increases with the rate at which the requests arrive from the users. I would have liked to see some performance numbers to illustrate the benefits of the proposed approach at various request rates.

Related Information



Nebeninhalt

Kontakt

Prof. Dr. Jens Teubner
Tel.: 0231 755-6481