|Review Comment: |
The author addressed the majority of my comments as well as my questions. The paper is definitely much clear now, and the examples precisely illustrate the power of the formalism. Overall, the reported work is very important to understand the behavior of existing engines, and is a valuable contribution for federated data management. Nevertheless, there are some definitions and properties that still need to be either explained in more detail or illustrated with more examples. First, just after Theorem 4.1 and Definition 4.6, the meaning of join-connected of BGPs and cross-site joins needs to be explained. Regarding to Theorem 5.5, please provide a decomposition of query LS5 that is agnostic and egalitarian. Give examples of unconstrained BGPs.
There are some errors in Tables 7 and 8 in pages 14 and 15, respectively. As indicated by Montoya et al. in 28 and 23, DEFENDER/ANAPSID assigns maximal stars that can be potentially executed by at least one endpoint. Thus, query LS5 is decomposed into three stars. The first one is composed of triple patterns in rows 1, 2, and 3, and it is assigned to Drugbank. The second star comprises triple pattern in row 4, and it is assigned to ChEBI and KEGG. Finally, triple patterns in rows 5 and 6 are assigned to ChEBI. This decomposition can be found at http://defender.ldc.usb.ve (Star-Shaped Groups with Union) and also presented in the paper by Montoya et al. COLD2012 Table 3.
Present references in increasing order, e.g., [18,17] => [17,18].
Make explicit the answer to my question Q3 in the conclusions of the paper.