From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id ; Tue, 8 May 2001 09:02:13 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id ; Tue, 8 May 2001 09:02:04 -0400 Received: from router-100M.swansea.linux.org.uk ([194.168.151.17]:25868 "EHLO the-village.bc.nu") by vger.kernel.org with ESMTP id ; Tue, 8 May 2001 09:01:52 -0400 Subject: Re: [RFC] Direct Sockets Support?? To: knicholson@corp.iready.com (Ken Nicholson) Date: Tue, 8 May 2001 14:04:45 +0100 (BST) Cc: alan@lxorguk.ukuu.org.uk ('Alan Cox'), Venkateshr@ami.com, pollard@tomcat.admin.navo.hpc.mil, linux-kernel@vger.kernel.org In-Reply-To: <034670D62D19D31180990090277A37B701811D72@mercury.corp.iready.com> from "Ken Nicholson" at May 07, 2001 06:55:08 PM X-Mailer: ELM [version 2.5 PL1] MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Message-Id: From: Alan Cox Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org > A couple of concerns I have: > * How to pin or pagelock the application buffer without > making a kernel transition. You need to pin them in advance. And pinning pages is _expensive_ so you dont want to keep pinning/unpinning pages > * Assuming the memory can be locked down, how can a list > of physical memory ranges be obtained (necessary to support > scatter/gather DMA? Is kiobufs suitable with it's page-alignment > constraints? If kiobufs will work, how can the kernel transition be > avoided? kiovecs will do that. It might be a little heavyweight but that should improve in 2.5 as we move to a slightly lighter model > WinSock Direct seems to address these concerns. These issues > become important at 1Gb and 10Gb speeds. 1Gbit - not really, 10Gbit yes