On Tue, May 11, 2010 at 11:29:08PM -0400, John Watlington wrote: > Reading the patch, it sounds like a better set of parameters > to try would be: > vm.dirty_background_ratio=10 > vm.dirty_bytes=90
dirty_bytes of 90 is rounded down to nearest page size; in this case zero. Instead, I suspect you mean: vm.dirty_background_ratio=10 vm.dirty_ratio=90 Which means that as a process writes, it will not suffer any impact until 10% of memory is filled with dirty buffers, then it will suffer pdflush contention, and then when 90% of memory is filled with dirty buffers it will face delays as writes will start writeback. With the 250Mb test I've been doing, this causes 8 seconds in dd, 147 seconds in sync, and a nice distribution of write latencies (largest samples 0.2s 0.1s 44ms 15ms 8ms, median 85us, smallest sample 80us). Oh I wish I had a storage activity LED on internal SD. ;-) -- James Cameron http://quozl.linux.org.au/ _______________________________________________ Devel mailing list [email protected] http://lists.laptop.org/listinfo/devel
