Jump label

Service navigation

Main navigation

You are here:

Main content

MXQuery With Hardware Acceleration — ICDE 2012 Reviews

Reviews for paper MXQuery With Hardware Acceleration, submitted to ICDE 2012.

Overall rating: accept

Reviewer 1

Overall Rating (Please be decisive toward a clear accept or reject choice)

Weak Accept

Presentation (Please indicate whether the paper is written in an acceptable manner for ICDE; badly written papers should be marked as unacceptable).

Good

Reviewer confidence (Please indicate how certain you are of the ratings given to the paper)

Medium

Summary of the paper (what is being proposed and in what context; short explanation of the recommendation). One paragraph

The paper proposes a demonstration of a hardware/software hybrid to speed up XQUery processing. An FPGA-based pre-filter is used to compute the XML projection for the given query, which can greatly reduce the workload of the subsequent XQuery processor. This promises to be an interesting demo, although some important technical aspects are still unclear.

Detailed comments

The general approach of using an FPGA as pre-filter to reduce the load on the XQuery processor is sound, and the indicated interactive demo setup is interesting. However some technical aspects are not explained sufficiently. First, can the FPGA handle arbitrary XML documents? FPGA based solutions usually do not like variable length input, how is this realized here? Also, how does the interface to the XQuery processor look like? Figure 2 seems to indicate that the FPGA really produces the filtered XML (i.e., the original XML document with the non-relevant nodes removed), but that seems to be particularly hard for an FPGA. After all, the root path is unbounded, and the FPGA cannot prune it easily in the presence of descendant steps.
Concerning the original motivation, if the parsing itself is on the critical path, why parse at all? The data could easily be stored in a pre-parsed or at least parser friendly binary format, which would dramatically reduce the parse time. And such a storage format would be completely transparent to the user, as the original XML document can be reconstruct from the binary XML format.
Finally, it would be nice to indicate the performance of the FPGA somewhere. I assume that it is bounded by wire speed (as shown by Q15), but it is never explicitly states if this is true for all queries.

Reviewer 2

Overall Rating (Please be decisive toward a clear accept or reject choice)

Accept

Presentation (Please indicate whether the paper is written in an acceptable manner for ICDE; badly written papers should be marked as unacceptable).

Excellent

Reviewer confidence (Please indicate how certain you are of the ratings given to the paper)

High

Summary of the paper (what is being proposed and in what context; short explanation of the recommendation). One paragraph

This demo paper describes MXQuery/H, an extension of the MXQuery engine with XML pre-filtering implemented in hardware using an FPGA. This improves the performance of XQuery execution by implementing efficient XML projections.
This is an interesting proposal for a demonstration that shows the benefits of hardware/software co-design. The proposal is well-written and clearly describes the novel system architecture, in which projection path information is dynamically installed on an FPGA. The benefit of the advocated approach is shown using a standard benchmark.

Detailed comments

* It would be interesting to see a comparison with the work described in reference [5].
* In the current system, the FPGA executes a linear filtering pipeline. Would it be possible to implement more complex pre-processing workflows on the FPGA?
What other operators in the XQuery model could be improved by doing this?
* What are the benefits of using a stream processor such as IBM's WSP as opposed to an FPGA?
* How is this different from other proposals for efficient stream/event processing in hardware, eg: Mohammad Sadoghi, Harsh Singh, and Hans-Arno Jacobsen. 2011. fpga- ToPSS: line-speed event processing on fpgas. In Proceedings of the 5th ACM international conference on Distributed event-based system (DEBS '11). ACM, New York, NY, USA

Reviewer 3

Overall Rating (Please be decisive toward a clear accept or reject choice)

Accept

Presentation (Please indicate whether the paper is written in an acceptable manner for ICDE; badly written papers should be marked as unacceptable).

Excellent

Reviewer confidence (Please indicate how certain you are of the ratings given to the paper)

Medium

Summary of the paper (what is being proposed and in what context; short explanation of the recommendation). One paragraph

This paper demonstrated MXQuery/H, a modified version of MXQuery that uses FPGA to speed up XML projection, which is a traditional problem in pre- processing XML documents preparing for more complicated XQuery later. It shows it's highly feasible to implement the dynamic query parts in an efficient way in FPGA to avoid recompilation/reloading overhead.

Detailed comments

In general, I like the idea of using FPGA for do expensive parsing/projections related tasks. My question is if similar idea can be applied to other serialization/deserialization formats, such as AVRO, or Protocal Buffer.

Related Information



Sub content

Contact

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