begin quoting David Brown as of Thu, Mar 27, 2008 at 07:47:11AM -0700: > On Sun, Mar 23, 2008 at 02:59:28PM -0700, Tracy R Reed wrote: > > >http://lwn.net/Articles/272534/ > > > >I really like this idea. Memory is plentiful and cheap. Good hardware is > >quite reliable these days and if you put a UPS behind it I could conceive > >of eventually putting a production database or some other seek-heavy > >application on such a storage system. > > I've been reading the feedback on LKML to this request. Someone asked him > the basic question (same one I asked) as to how this differs from just > mounting the filesystem normally with auto-flushing turned off. > > Basically, he's too confrontational for his patches to be accepted. He > doesn't listen to people suggestions.
I got that from the 'you trust' comment in the article. > The basic concensus seems to be that the technique he uses to write back to > disk really isn't all that useful. If there is a failure, you're going to > have to restore it from some kind of backup. Hardware never fails. Programs never crash. Breakers never trip. Three year olds never stick a stick into the fan of the power supply to hear the funny noise. > One suggestion was to just mount a normal ramdisk and have a user-space > program use dnotify to synchronize with a disk filesystem. The disk > filesystem would have normal journalling and such. > > Now if someone could just figure out how to not require every directory to > be watched explicitly with dnotify, so that it could be useful for > something like this. http://inotify-tools.sourceforge.net/ inotifywait This command simply blocks for inotify events, making it appropriate for use in shell scripts. It can watch any set of files and directories, and can recursively watch entire directory trees. -- Make a wish, get a wish. Stewart Stremler -- [email protected] http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
