> Is it that nodes store redundant data to take over the responsibilities of a next door neighbor node going down?
In my understanding, this extra-successor scheme has little (or nothing) to do with data redundancy. Instead, it deals with what you can call meta-data redundancy: instead of making a node knowing how to contact one other node, let's make it store more information so that it knows how to contact more nodes (successors). This way, the probability of a node having all neighbors experiencing a failure can be brought low.
> I'm also somewhat confused why they are strict about going counterclockwise. If they know if the desired node is higher or lower in the ring, they could easily implement searching in both directions and reduce the average search distance. This maybe doesn't matter because of the exponential jumps in "known" node distances...but it seems that it would be a cheap efficiency gain on some level.
I can only speculate that this is because the DHT is usually presented as a logical overlay. So the hops there are logical hops, not necessarily representative of actual network distance.
In my understanding, this extra-successor scheme has little (or nothing) to do with data redundancy. Instead, it deals with what you can call meta-data redundancy: instead of making a node knowing how to contact one other node, let's make it store more information so that it knows how to contact more nodes (successors). This way, the probability of a node having all neighbors experiencing a failure can be brought low.
> I'm also somewhat confused why they are strict about going counterclockwise. If they know if the desired node is higher or lower in the ring, they could easily implement searching in both directions and reduce the average search distance. This maybe doesn't matter because of the exponential jumps in "known" node distances...but it seems that it would be a cheap efficiency gain on some level.
I can only speculate that this is because the DHT is usually presented as a logical overlay. So the hops there are logical hops, not necessarily representative of actual network distance.