From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id ; Thu, 2 May 2002 16:15:08 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id ; Thu, 2 May 2002 16:15:06 -0400 Received: from host.greatconnect.com ([209.239.40.135]:20228 "EHLO host.greatconnect.com") by vger.kernel.org with ESMTP id ; Thu, 2 May 2002 16:14:59 -0400 Message-ID: <3CD19D16.7070605@rackable.com> Date: Thu, 02 May 2002 13:09:58 -0700 From: Samuel Flory User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.0rc1) Gecko/20020417 X-Accept-Language: en-us, en MIME-Version: 1.0 To: Roy Sigurd Karlsbakk CC: =?ISO-8859-1?Q?Jakob_=D8stergaard?= , Martin Dalecki , Pavel Machek , linux-kernel@vger.kernel.org Subject: Re: IDE hotplug support? In-Reply-To: <20020426152943.A413@toy.ucw.cz> <3CD18318.7060407@evision-ventures.com> <20020502215833.V31556@unthought.net> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 8bit Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Why not just grab a pair of 8 port 3ware cards? Run raid 5 on each card, and throw 0 or linear via the md driver on top? Jakob Østergaard wrote: >On Thu, May 02, 2002 at 08:19:04PM +0200, Martin Dalecki wrote: >... > > >>15 drives == 16 interfaces == 8 channels == 4 controllers >>with primary and secondary channel. >> >> > >Usually using both master and slave on an IDE channel spells disaster >performance wise, and I would be surprised if the hotplug stuff worked >with this as well... > > > >>He will have groups of about 4 drives on each channel wich >>serialize each other due to excessive IRQ line sharing and >>master slave issues. >> >> > >Use 8 controllers for the 15 (16) drives. > > > >>8 x 130MBy/s >>>> PCI bus throughput... I would rather recommend >>a classical RAID controller card for this kind of >>setup. >> >> > >Because RAID controllers do not use the PCI bus ??? ;) > >The bus-overhead on RAID-5 is not too bad unless you specifically construct >a workload to make it so (writes-only, scattered so that the kernel cannot >cache stripes to avoid read-in for parity calculation). > >Sure, the PCI bus will be a bottleneck, and PCI overhead alone will decrease >the real-world performance to somewhere below the theoretical PCI bandwidth >limitations, but don't let this blind you - 100 MB/sec sustained transfers >can still be "good enough" for many people. > >By the way, has anyone tried such larger multi-controller setups, and tested >the bandwidth in configurations with multiple PCI busses on the board, versus a >single PCI bus ? > > >