TalkChess.com

[Vincent Diepeveen]

CPU's that don't cause a cutoff at this node yet do unsubscribe themselves from this search and basically are idle, i called that cpu's calling AbortMe().

CPU's that cause either a research (PV) or fail high at this node, causing all other cpu's to get aborted, that's what i described as an AbortFailHigh().

CPU's that call AbortMe() yet then are idle because they own the specific 'recursionstack' and which activelly go search THEMSELVE higher in the search tree (so not lower which is what Crafty is doing) and/or at other CPU's stealing work there, that's basically a DTS feature as invented by Bob.
From what i understood - long time ago - in DTS cpu's actively search themselves for a job.

It does give a good speedup. So from algorithmic viewpoint it's a good thing.

Note i'm not a fan of this at modern hardware as it doesn't scale IMHO, yet YMMV there as Diep of course is in a bigger luxury position there as it doesn't have a recursive search at all, so it doesn't have this problem.

It's not clear to me whether you can claim a new name for this. Have to ask Bob careful for that - again with Diep i don't need a solution like this as it's in the first place already having a superior design there.

At the cluster though i need creative stuff to solve it all as it has more limitations of course with remote nodes. So i'm already looking there to memory migration, yet again in a manner that idle cpu's do not search themselves for a job

The guy who really has been busy with that is Bob.