Newsgroups: comp.parallel
From: vanvoors@boreas.src.honeywell.com (Brian VanVoorst)
Subject: Selection problem (2nd try)
Organization: Honeywell Technology Center
Date: 21 Dec 1995 05:16:12 GMT
Message-ID: <4b99q5$ea7@moon.src.honeywell.com>

(I didn't see my first post posted so I am trying again.)

	[The Mod. apologizes on this: the closing of the US Federal
	govt. complicates moderation at the moment along with a couple
	of small technical problems.  --enm]

I am looking for code, info, papers on the parallel selection problem.  (Code
prefered).  The problem is you have N numbers distributed over P processors,
and you want to find the kth one.

Some people have suggested to me there might be an O(N) answer but have been 
unable to describe what it would be.  On the other hand I have a working  
O (n/p log n/p) answer.  This requires each node to sort their own numbers 
O (n/p log n/p) and then at most (log n/p + 1) gather / scatter messages are 
needed to find the kth one.

Any pointers to sample codes or papers would be appreciated.

Thanks
Brian


