Figure 2a |
Project on a range of input data sizes. |
Figure 2b |
Join on data sets of different structure. |
Figure 3a |
PageRank on the Orkut follower graph. |
Figure 3b |
PageRank on the Twitter follower graph. |
Figure 7 |
Comparison of hand-coded jobs and Musketeer on TPC-H query 17 (business analytics). |
Figure 8a |
Comparison of hand-coded baselines and Musketeer jobs for PageRank (Orkut). |
Figure 8b |
Comparison of hand-coded baselines and Musketeer jobs for PageRank (Twitter). |
Figure 8c |
Resource efficiency of different back-end execution engines for PageRank (Twitter). |
Figure 9 |
Cross-community PageRank on the LiveJournal graph and a synthetic graph. |
Figure 10 |
Movie recommendation engine using the NetFlix data set. |
Figure 11 |
Overhead of Musketeer-generated code for PageRank over hand-optimized baselines. |
Figure 12a |
Benefits of Musketeer's operator merging for the "top-shopper" workflow. |
Figure 12b |
Benefits of Musketeer's operator merging for the cross-community PageRank workflow. |
Figure 13 |
Decision time of Musketeer's scheduling algorithms as a function of workflow size. |
Figure 14 |
Decision quality of Musketeer's automatic back-end choices. |
Figure 15a |
Automated mapping choice correctness for single source shortest path (SSSP). |
Figure 15b |
Automated mapping choice correctness for k-means clustering. |