I think the iterate until convergence optimization can be implemented using the frontier by making the frontier contain vertices which haven't converged in each iteration. This would then be equivalent to round robin scheduler in graphlab, where the recently updated elements would be available for next iteration.
lya
I think we can modify the COND here to shrink the result of EDGEMAP is some vertices are converged. It is very similar to the dynamic scheduler in GraphLab, i.e. using signaling.
I think the iterate until convergence optimization can be implemented using the frontier by making the frontier contain vertices which haven't converged in each iteration. This would then be equivalent to round robin scheduler in graphlab, where the recently updated elements would be available for next iteration.
I think we can modify the COND here to shrink the result of EDGEMAP is some vertices are converged. It is very similar to the dynamic scheduler in GraphLab, i.e. using signaling.