![]() This is true for a small and medium number of PoPs. This is indeed true, but because our problem space is so small (there are less than 100K 7th level s2 cells ) we can just brute-force through it and get a definitively optimal result instead of the one that can get stuck on a local optimum.Īnycast-based loadbalancing is mostly optimal but it behaves poorly on high percentiles. Some of you may see that the problem here that can be solved by more sophisticated methods like Gradient Descent or Bayesian Optimization. In our case we try to overcompensate for the effects of latency on the TCP throughput. As for the loss function to determine the score of each placement one can use something standard like L1 or L2 loss. Doing exhaustive search to find the “best” location for the new PoPīy “population” one can use pretty much any metric we want to optimize, for example total number of people in the area, or number of existing/potential users.Computing the distance to the nearest PoP for all the regions weighted by “population”.Splitting the Earth into 7th level s2 regions.We try to alternate new PoP placement between selecting the most advantageous PoP for the existing and potential Dropbox users.Ī tiny script helps us brute-force the problem by: what location will benefit Dropbox users better, Vienna or Warsaw? The problem persists as the number of PoPs grows: e.g. ![]() Even with a small number of PoPs without assistive software it may be challenging to choose between, for example, a PoP in Brazil and a PoP in Australia. The current PoP selection procedure is human guided but algorithm-assisted. The process of PoP selection, which was easy at first, now becomes more and more complicated: we need to consider backbone capacity, peering connectivity, submarine cables, but most importantly the location with respect to all the other PoPs we have.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |