Reviews for paper Why Off-The-Shelf RDBMSs are Better at XPath Than You Might Think, submitted to ICDE 2007.
Overall rating: reject
Yes, Probably
Yes
Weak reject (Not very innovative)
Weak reject (Fair)
Weak reject (Fair)
Accept (Good)
Weak reject (Probably reject)
High (I know this area well)
20%
No
The paper uses a relational encoding of xml data and a commericial database to evaluate XPath expressions. The key contribution over similar papers on the topic is the placing 'level' at the front of the key to create a partitioned B-Tree. The paper does not consider value predicates in the XPath expressions. The paper also does include the cost of the retrieval of large portions of the document to the client as well as processing over transient XML fragments caused by construction in XQuery.
The relational database vendors have recently described their XML indexing techniques. The XML indexes in IBM DB2 and Microsoft SQLServer both have a version that includes a path at the front of their key, which effectively creates similar partitioned B-Trees. A comparison with these systems would be very useful.
page 2: Isn't post - pre = size? (vs size - level)
page 2: The big problem with the range condition, pre(c) < pre(v) <= pre(c) + size(c), is that index is only useful in one direction: we can find all the v's given a c, but we cannot find all the c's from a v very efficiently.
Yes
Yes, Definitely
Yes
Accept (Innovative)
Weak accept (Good)
Weak accept (Good)
Strong accept (Excellent)
Weak accept (Could go either way)
Medium (Moderately confident)
Quite a few - XML storage and processing are popular topics.
No
This paper looks at how indexing and query processing and rewrite techniques employed in modern commercial database systems - using DB2 as the poster child for a modern commercial system - can be directly applied to XPath processing over untyped XML. I liked the approach - no modifications allowed, what can be done, why does it work, how does it relate to more XML-oriented ideas? - to the problem. The paper is practically-oriented and very well written. The main weakness of the paper is that it is somewhat narrowly focused on path expressions, so the generality / overall usefulness of the study is limited in that regard.
Nice job! One thing was unclear to me - related to B-tree partitioning and your experiments - are you recommending to use a (level, pre) index **instead** of just a pre index, or **in addition** to the pre index? I assumed the latter. (Since you are not studying updates, presumably you simply create all the indexes you need for the various path queries of interest, right?)
Yes
Yes, Definitely
Yes
Weak reject (Not very innovative)
Weak accept (Good)
Weak accept (Good)
Weak reject (Fair)
Weak accept (Could go either way)
Medium (Moderately confident)
50
No
The paper deals with processing XPath where the XML data is relationally encoded. It makes some interesting observations about how indexing can accommodate the efficient processing of an Xpath query which is expressed as an SQL query.
This paper fits well to the Industrial track as it deals with using existing relational systems, without alterations, for efficient XPath processing. However, the experimentation presented is not comprehensive.
- Is the 'child property' equation '(CHILD)'?
- Partitioned B-trees and ORDPATH: this paragraph is too abbreviated to be of use.
- 'a implementation' --> 'an implementation'.
'executed as single record scan' - What does this mean exactly? Just obtaining the 1st record from this scan?
- Figures 8 + 9: The confusing part is that you eventually select the child and not the parent, a bit convoluted.
Section 4: the rewriting of (1) into (20 is not entirely clear.
- Section 5 is mostly brainstorming.
Yes