From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1756136Ab1FJR1O (ORCPT ); Fri, 10 Jun 2011 13:27:14 -0400 Received: from fmailhost03.isp.att.net ([204.127.217.103]:59888 "EHLO fmailhost03.isp.att.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752889Ab1FJR1M (ORCPT ); Fri, 10 Jun 2011 13:27:12 -0400 X-Greylist: delayed 875 seconds by postgrey-1.27 at vger.kernel.org; Fri, 10 Jun 2011 13:27:12 EDT X-Originating-IP: [70.241.89.83] Date: Fri, 10 Jun 2011 12:13:17 -0500 From: Robert Redelmeier To: LKML Cc: Robert Redelmeier Subject: NEPOTISM_BONUS yield_to_child() ? Message-ID: <20110610171317.GA15253@jim.domain_not_set.invalid> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.4.2.3i Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org [YAFR - Yet Another Feature Request] Multi-threaded/tasking performance is entirely dependent on the OS scheduler. I searched /usr/src/linux/Documentation/scheduler/* and found no mention of parent or child processes. Parent- and child-processes are interconnected, and for maximum [SMP] performance, these connections should be recognized _optimally._ For example, on some [desktop] systems, an exec() syscall would run unblocked, an exit() would return the rest of the timeslice to the parent, yield() would go to child (parent?), etc. Other servers might choose the exact opposite. This connectivity could be implemented by a tuneable scheduler param: NEPOTISM_BONUS = int where positive values would influence the scheduler to run parents/children/uncles/nieces and negative values would cause the scheduler to prefer strangers [unrelated procs]. -- Robert in Houston