On Wed, Jan 28, 2015 at 01:58:55PM +0900, Minchan Kim wrote: > On Wed, Jan 28, 2015 at 01:50:28PM +0900, Sergey Senozhatsky wrote: > > On (01/28/15 13:07), Sergey Senozhatsky wrote: > > > On (01/28/15 12:53), Sergey Senozhatsky wrote: > > > > > So, I want to go with srcu. Do you agree? or another suggestion? > > > > > > > > yes, I think we need to take a second look on srcu approach. > > > > > > > > > > ... or we can ask lockdep to stop false alarming us and leave it as is. > > > I wouldn't say that ->init_lock is so hard to understand. > > > just as an option. > > > > > > > so... returning back to barriers performance implications. > > > > x86_64, lzo, 4 comp streams, 2G zram, ext4, mount -o rw,relatime,data=ordered > > > > ./iozone -t 3 -R -r 16K -s 60M -I +Z > > > > test base srcu > > " Initial write " 1299639.75 1277621.03 > > " Rewrite " 2139387.50 2004663.94 > > " Read " 6193415.00 5091000.00 > > " Re-read " 6199050.38 4814297.88 > > " Reverse Read " 4693868.88 4367201.75 > > " Stride read " 4470633.75 4247550.00 > > " Random read " 5115339.50 4517352.75 > > " Mixed workload " 4340747.06 3880517.31 > > " Random write " 1982369.75 1892456.25 > > " Pwrite " 1352550.22 1248667.78 > > " Pread " 2853150.06 2445154.41 > > " Fwrite " 2367397.81 2262384.56 > > " Fread " 8100746.50 7578071.75 > > > > not good. > > > > Oops, I never thought it could make mesurable performance. > I will investigate it. > > Thanks a lot, Sergey! Sergey, the data is consistent for repeated work? I tested it with dd on /dev/zram0 without any FS on my KVM and I cannot see any measureable performance gap. Hmm, I will try it on real machine. -- Kind regards, Minchan Kim -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>