From mboxrd@z Thu Jan 1 00:00:00 1970 From: Joel Soete Subject: Re: [parisc-linux] ext3 perf patch#4 Date: Sun, 27 Mar 2005 13:49:06 +0000 Message-ID: <4246B9D2.7070906@tiscali.be> References: <20050326103540.GA31557@colo.lackof.org> <20050327011849.GD9287@colo.lackof.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii; format=flowed Cc: parisc-linux@lists.parisc-linux.org To: Grant Grundler Return-Path: In-Reply-To: <20050327011849.GD9287@colo.lackof.org> List-Id: parisc-linux developers list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: parisc-linux-bounces@lists.parisc-linux.org Grant Grundler wrote: > On Sat, Mar 26, 2005 at 03:35:40AM -0700, Grant Grundler wrote: > >>James Bottomley in a private conversation suggested our >>ext2_test_bit() and ext2_find_next_zero_bit() might not >>be counting bits the same way in the bitmap. > > > James, kudos and thank you. > > With appended patch, I'm getting reasonable ext3 performance. > Can someone please test this patch in a 32-bit kernel? > Obviously ;-) > I'm comfortable this patch works for 64-bit since I've been building > kernels and running rsync across 3 disks with no problems. > I *think* the code is 32-bit clean as well, but haven't tested it. > on my c110 no pb to boot :-) some rsync test (on non mirored slices on 2 different disk but on the same scsi ctrlr ncr53c720): # time rsync -a /Debian-apt/SRC/linux-2.6.12-rc1-pa4-050327 /Sources/CVS-tst real 3m34.512s user 1m0.730s sys 1m6.440s # du -sk /Debian-apt/SRC/linux-2.6.12-rc1-pa4-050327 256616 /Debian-apt/SRC/linux-2.6.12-rc1-pa4-050327 ... > Simple test results follow and then the patch. > # bdf Filesystem 1K-blocks Used Available Use% Mounted on /dev/md2 1692128 1194620 411552 75% / tmpfs 256300 0 256300 0% /dev/shm /dev/md0 123699 112645 4667 97% /boot /dev/md3 247407 82537 152096 36% /var /dev/md4 123635 4129 113122 4% /tmp /dev/md5 123635 15749 101502 14% /home /dev/sdb9 1513656 968940 529340 65% /Debian-apt /dev/sda9 1513656 1011696 425068 71% /Sources /dev/sdc5 1692220 913028 693232 57% /chroot /dev/sdc3 123767 8728 108649 8% /chroot/boot /dev/sdc6 247511 107636 127096 46% /chroot/var /dev/sdc7 123735 4130 113216 4% /chroot/tmp /dev/sdc8 123735 4144 113202 4% /chroot/home /dev/sdc9 1513656 473444 1024836 32% /chroot/Develop ... > grundler@riot:/$ lsscsi Not yet lsscsi > > riot:/mnt# time tar xjf linux-2.6.12-rc1-pa3.tar.bz2 > I prefer simply tar and so: # cd /Debian-apt/SRC # time tar -cslpf /chroot/Develop/linux-2.6.12-rc1-pa4.tar2 linux-2.6.12-rc1-pa4-050327 real 0m38.487s user 0m3.690s sys 0m18.650s a sample iostat too :-) avg-cpu: %user %nice %sys %iowait %idle 15.70 0.00 47.00 37.30 0.00 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 2.50 0.00 19.60 0 196 sdb 297.30 5906.40 89.20 59064 892 sdc 15.10 0.00 7712.80 0 77128 md0 0.00 0.00 0.00 0 0 md5 0.00 0.00 0.00 0 0 md4 0.00 0.00 0.00 0 0 md3 0.20 0.00 0.40 0 4 md2 2.00 0.00 16.00 0 160 md1 0.00 0.00 0.00 0 0 # cd /chroot/Develop # time tar -xslpf linux-2.6.12-rc1-pa4.tar2 real 1m32.901s user 0m4.880s sys 0m30.000s another iostat sample: avg-cpu: %user %nice %sys %iowait %idle 13.19 0.00 39.66 47.15 0.00 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 0.00 0.00 0.00 0 0 sdb 0.00 0.00 0.00 0 0 sdc 173.33 4542.66 6186.61 45472 61928 md0 0.00 0.00 0.00 0 0 md5 0.00 0.00 0.00 0 0 md4 0.00 0.00 0.00 0 0 md3 0.00 0.00 0.00 0 0 md2 0.00 0.00 0.00 0 0 md1 0.00 0.00 0.00 0 0 still have to test on a quicker system as the b2k if that could solve the bug I reproduce on 32bit kernel (cf patch log message:"... I suspect the real problem is ffz() wants an unsigned long and was getting garbage in the top half of the unsigned int. Not confirmed but that's what I suspect.) Thanks, Joel _______________________________________________ parisc-linux mailing list parisc-linux@lists.parisc-linux.org http://lists.parisc-linux.org/mailman/listinfo/parisc-linux