Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP Posting-Version: version B 2.10.1 6/24/83; site whuxk.UUCP Path: utzoo!watmath!clyde!burl!hou3c!hocda!houxm!whuxl!whuxk!dkit From: dkit@whuxk.UUCP (LIPINSKI) Newsgroups: net.lang Subject: Re: Fastest parallel Sort Message-ID: <62@whuxk.UUCP> Date: Thu, 21-Jun-84 23:57:17 EDT Article-I.D.: whuxk.62 Posted: Thu Jun 21 23:57:17 1984 Date-Received: Fri, 22-Jun-84 07:35:28 EDT References: <503@spuxll.UUCP> Organization: Bell Labs, Whippany, N.J. Lines: 9 The fastest parallel sort has recently been proved to be O(log n), which improves upon Batcher's O((log n)^2). It's in a paper by Komlos, Ajtai, and Sar??? (sp) [Stanford &/| UCSD], and it uses expander graphs and a linear nearsort algorithm. The problem is that the linear coefficient is huge (10^20?), so it's no good to actually implement it; it just proved it can be done in O(log n). Michael Baldwin {harpo,ihnp4}!whuxg!mike