One reason that we can choose the scheduling policy is that most of graph-based algorithm will converge, thus the asynchronous execution will not influence the correctness of the algorithm.
nemo
What should be the best scheduling policy for PageRank to ensure correctness?
apadwekar
Users of GraphLab have a better sense of what type of consistency is necessary, so GraphLab allows application developers to specify the exact scheduling policy.
For PageRank, synchronous would definitely ensure correctness. However, it is unclear if the others will ensure exact correctness it is clear they will converge to approximately the same value (another pagerank algorithm starts with all the weight on one vertex and distributes it from there).
dmerigou
For instance, in the 2D grid solver example used in the beginning of the course, synchronous or graph coloring schedules would work.
One reason that we can choose the scheduling policy is that most of graph-based algorithm will converge, thus the asynchronous execution will not influence the correctness of the algorithm.
What should be the best scheduling policy for PageRank to ensure correctness?
Users of GraphLab have a better sense of what type of consistency is necessary, so GraphLab allows application developers to specify the exact scheduling policy.
For PageRank, synchronous would definitely ensure correctness. However, it is unclear if the others will ensure exact correctness it is clear they will converge to approximately the same value (another pagerank algorithm starts with all the weight on one vertex and distributes it from there).
For instance, in the 2D grid solver example used in the beginning of the course, synchronous or graph coloring schedules would work.