From mboxrd@z Thu Jan 1 00:00:00 1970 From: Douglas Gilbert Subject: Re: [patch for playing] 2.5.65 patch to support > 256 disks Date: Tue, 25 Mar 2003 08:22:51 +1000 Sender: linux-scsi-owner@vger.kernel.org Message-ID: <3E7F853B.1020603@torque.net> References: <200303211056.04060.pbadari@us.ibm.com> <3E7C4D05.2030500@torque.net> <20030322040550.0b8baeec.akpm@digeo.com> <200303241332.56996.pbadari@us.ibm.com> Reply-To: dougg@torque.net Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit Return-path: List-Id: linux-scsi@vger.kernel.org To: Badari Pulavarty Cc: Andrew Morton , linux-kernel@vger.kernel.org, linux-scsi@vger.kernel.org, Jens Axboe Badari Pulavarty wrote: > On Saturday 22 March 2003 04:05 am, Andrew Morton wrote: > >>OK, thanks. So with 48 disks you've lost five megabytes to blkdev_requests >>and deadline_drq objects. With 4000 disks, you're toast. That's enough >>request structures to put 200 gigabytes of memory under I/O ;) >> >>We need to make the request structures dymanically allocated for other >>reasons (which I cannot immediately remember) but it didn't happen. I >>guess we have some motivation now. > > > Here is the list of slab caches which consumed more than 1 MB > in the process of inserting 4000 disks. > > #insmod scsi_debug.ko add_host=4 num_devs=1000 > > deadline_drq before:1280 after:1025420 diff:1024140 size:64 incr:65544960 > blkdev_requests before:1280 after:1025400 diff:1024120 size:156 incr:159762720 > > * deadline_drq, blkdev_requests consumed almost 80 MB. We need to fix this. > > inode_cache before:700 after:140770 diff:140070 size:364 incr:50985480 > dentry_cache before:4977 after:145061 diff:140084 size:172 incr:24094448 > > * inode cache increased by 50 MB, dentry cache 24 MB. > It looks like we cached 140,000 inodes. I wonder why ? > Badari, What number do you get from # cd /sys; du -a | wc when you have 4000 disks? With two disks the count on my system is 528. Also scsi_debug should use only 8 MB (default) for simulated storage shared between all pseudo disks (i.e. not 8 MB per simulated host). Doug Gilbert