I originally installed Apache Superset on the same AWS node that I have the coordinator and overlord processes running on. Superset ran ok. Then I moved it to its own utility node to see if it would be faster. Turns out it is much slower (queries are now timing out). Superset does not suggest a specific node for installment.
In theory it shouldn't matter if nodes are within the same network switch. In fact the size of the result returned by the coordinator is very very small (maybe couple of 100 bytes), in addition to that the bulk of the data is coming from the broker (i will worry about this link), therefore this behavior seems strange to me. I am not sure how amazon allocates nodes but i will first checkout if all your druid cluster is within the same switch (reasonable network latency). Giving such circumstances i will collocate superset with the borker node instead. Final thought are you sure the timeout is related to network issue and not the query taking too much time ?