Parallel search once more

[bob]

I tested these positions and they cause significant problems for fixed-depth testing using parallel search. I had problems with at least 10-15% of them. What seemed to be happening was that there is some critical path for a position that is very hard to find (probably why you noticed the disagreement between two strong programs for 4 consecutive plies). The parallel search sometimes is able to "sniff" out part of the path and the search tends to explode when that happens.

On one attempt a 30 ply search might take 90 seconds, the next run might see the 29 ply search run for 1800 seconds. That certainly makes fixed-depth testing painful. Quite a few of them are useful, but some just don't work well for me.