Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: For the Record: Effects of Larger Hash Tables

Author: Gerd Isenberg

Date: 11:20:57 12/03/05

Go up one level in this thread


On December 03, 2005 at 14:05:48, Gerd Isenberg wrote:

>On December 03, 2005 at 11:56:42, Andreas Guettinger wrote:
>
>>On December 03, 2005 at 11:01:46, Paul Jacobean Sacral wrote:
>>
>>>I would appreciate a couple of clarifying remarks as well, because this is a
>>>topic that's difficult to understand if you are not a progammer. Bacically, I
>>>was studying explanations of this in the past but didnt't understand all of it,
>>>and also do not remember all of it.
>>>
>>>My question is:
>>>
>>>How come that some solving times of test positions are worse (longer) with
>>>bigger hash tables, than with smaller hash tables?
>>>
>>>Yours truly Paul J. Sacral
>>
>>Can you give an example?
>>The size of the hashtable should not make a considerable difference in solving
>>time, except if the engine clears the hashtable in analysis mode at the
>>beginning of the search, which could take 1 or 2s on slow hardware. (Note during
>>normal gameplay hastables usually don't get cleared.)
>>Per position (in the search tree) it takes normally 1 (written ONE) probe per
>>hashtable, doesn't matter if the hashtable is 1Mb or 1Gb.
>>As a banal example, if you have a file register in your office, and you want to
>>lookup file no. 56, it doesn't mater if you have 100 or 1000 files stored, you
>>just walk to the shelf and take file number 56.
>>
>>regards
>>Andy
>
>With changed table size you obviously map positions to other entries due to
>hasIndex ::= someHashkey % tablesize. Two positions with disjoint slots with
>some table size may share one slot with a bigger hash size. This is enough to
>explain completely different search behaviour - and for some patological cases
>even a longer solving time.
>
>Gerd

Ok, the above statement is only true if table size if not power of two and you
have hashIndex ::= someHashkey & (2**log2tblsize - 1).
So different replacements of positions which map to same entries is the main
cause to change search behaviour with the mentioned patological cases.
Isn't "hash luck" is a common term?

Gerd



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.