hmm...i always assumed the graph subproblems are solved in parallel in the same process/instance like go-process , didn't realized they get distributed across multiple nodes/machines...but for a massive scale(that u guys aiming for) it definitely makes sense...good to know the insights...