On Mon, Apr 28, 2014 at 05:53:44PM +0200, Bernhard Voelker wrote: > On 04/28/2014 05:01 PM, Pádraig Brady wrote: > >Note the kernel may cache lots more than 4M before it > >even starts writing out to the device. > >I think there maybe some very recent improvements > >to the kernel to address at least the excessive > >caching part of the issue, though if there are > >large pending writes to slow devices the kernel > >can still block for a long time. > > Okay, that sounds reasonable. > I.e., in average the blocking time could be smaller, > but if many writes are pending to be flushed to disk > as in this case, then ~1min is reasonable for a slow > device. > > Thanks & have a nice day, > Berny
This is in fact the case here. If i wait for some seconds before killing it, it takes some (long) time to finish and after that states that it copied about 500MB to 1GB. So I guess it is in fact a hughe buffer to ram that is implicitly used by the OS. And of course if you have to wait for the pending 500MB to 1GB it takes a while to finish. The solution in my case is in fact to use oflag=direct. This way no caching (at least not that much caching) is done. Thanks, Sebastian
signature.asc
Description: Digital signature
