Gregory Todd Williams SPARQL BGP Optimization Presentation Jesse Weaver 2

From Semantic Portal Wiki

• Question is for the Presentation: Gregory Todd Williams SPARQL BGP Optimization Presentation
• Question is asked by: Jesse Weaver
• The Question is: Why are the selected queries inconsistent between figures/tables? (See question page for more details.)
• Answer: Figure 4 shows four queries which the paper suggests are the most interesting queries for BGP optimization (those with more than two triple patterns, etc.). I'm not sure why figure 5 replaces query 12 with query 4, but in general, the same "interesting" notion probably applies. Figure 6 likely uses query 2 for analyzing the plan space because it has the largest plan space of all the LUBM queries (and, thus, more potential for a naive query plan to execute very poorly). Table 1 shows all queries with at least three triple patterns except query 9 (whose analysis was prevented due to the plan space analysis taking greater than 2 weeks).

The reported data seem somewhat scattered.

Figure 4 reports query performance for LUBM queries 1, 2, 9, and 12; why only these four?

Figure 5 reports sampling performance for LUBM queries 1, 2, 4, and 9; why only these four?

Figure 6 reports plan space for only LUBM query 2; why just the one query? (Granted, a new chart for each query would have to be used and would take up a lot of space, but it seems that at least one or two more queries would give us a more well-rounded view of plan space... especially given that LUBM query 2 returns 0 results on the LUBM(1,0) dataset.)

Table 1 compares minimum, maximum, and ARQ/PFJ times for LUBM queries 2, 4, 7, 8, and 12; why only these five? Also, Table 1 claims to be for "the LUBM queries with at least three triple patterns," yet LUBM query 9 (which was used in Figure 4 and Figure 5) contains six triple patterns and is not included in Table 1.

Why are the selected queries inconsistent between figures/tables?