Hi again, I tested HDD with iozone on both linux and OpenBSD, with softdeps, disk cache between 5 and 50%, and namei augmented ( following tha faq about hdd performance ) And still get the same poor performance if compared with linux.
17Mb/s linux, and openbsd 5Mb/s Any other suggestion? 2007/7/19, Tang Tse <[EMAIL PROTECTED]>: > > Thanks for the answear > > Efectively, i have no softdep enabled on disk. > > Is possible to enable more ram for disk cache than 5%? > > > > 2007/7/18, francisco roque < [EMAIL PROTECTED]>: > > > > On Wed, 18 Jul 2007, Tang Tse wrote: > > > > > Are you using the same part of the disk for both tests? > > > - Yes on both, is an old scsi controller but supported ( I checked the > > HLC ) > > > > > > Is the OpenBSD fs using softdep? > > > - How can i check this? > > > > `mount -v` will report 'softdep' for the filesystem in question if it is > > > > enabled. I believe you need to manually enable it and doing so should > > increase general OpenBSD disk performance, not sure for this specific > > test. > > > > > > > > > > What is the amount of memory in the machine? > > > - 2Gb > > > > > > How many runs is this the average of? > > > - On linux in the same conditions ( clean install, dd from the same > > > partition, etc.. ) i get 17MB/s > > > > > > Is the input_file created immediately before the test? > > > Yes on both > > > > This, the amount of memory available, and the size of the file probably > > causes the biggest difference. IIRC, linux uses almost all available > > memory as filesystem cache, but OpenBSD uses 5% by default. In this > > case, the 1GB file will have been placed in fs cache when created on > > linux, but not on OpenBSD since it wouldn't fit. > > > > In other words, on linux you are testing reading from memory and writing > > to disk, but in OpenBSD you are testing reading from disk and writing to > > > > disk. > > > > A couple ways around this would be to either test files > 2GB or to > > create > > the file, umount the partition, mount it, then run dd. If you run > > IOzone > > instead of dd, the results can show you the performance of each system > > both when files fit in the memory cache and once it's out. What > > performance characteristics are best for your app is for you to decide > > (and often a bit beyond anything dd proves). > > > > > > > Is the machine running other processes at the same time? > > > Yes, I made a new instalation with minimum but i don't disable > > anything. I > > > want to test it on daily conditions to see the real performance. > > > > While that's nice and generally safe, you'll also need to be aware of > > what's happening in the background that might be time dependant and > > different between the two machines. E.g. testing OpenBSD on Saturday > > early morning while it's updating the locate database will show > > different > > results versus any other time of day. > > > > > > > > > > Thanks. > > > > > > > > Good luck, > > > > -f > > http://www.blackant.net/
