On Wed, Mar 15, 2006 at 11:10:48 -0800, antonyn@xxxxxxxxxxxxxxxxxxx wrote: > Hi Bruno, > > On Wed, 15 Mar 2006, Bruno Wolff III wrote: > > >On Wed, Mar 15, 2006 at 07:49:59 -0600, > > Les Mikesell <lesmikesell@xxxxxxxxx> wrote: > >>Doing a cat of the raw device to /dev/null once in a while will > >>give smartd a chance to report errors. The drive should even > >>internally remap a certain number of 'soft' errors transparently > >>if it is able to recover the data with retries but again it > >>will only do that if you access the bad spot. > > > >You should be using long self tests to do surface scans of the disk. > > Are you saying Les's idea of catting the raw device to /dev/null is not > sufficient to do a surface scan test? If it isn't, what is your > recommendation? Probably it is overkill. Instead of having the device do its own internal tests you are blowing out your cache. There also can be issues with the reads being optimized away. Some of the disk blocks could be in the OS cache when you start reading the disk. This probably isn't a big problem in practice. The self test may also test for other problems with the drive that reading wouldn't. It might not be a bad idea to do the the reads when you know the performance impact won't be an issue for you, but that should be in addition to using smartd to schedule long self tests.
