small files - linux kernel tree - 13381456448 bytes / 105601 files
big files - 10x linux .git pack - 13062560965 bytes

test1 - novafs
test2 - novafs - with clwb instead of movnti
test3 - ext4 on pmem with dax
test4 - ext4 on pmem
test5 - ext4 on pmem with btt
test6 - tmpfs
test7 - ext4 on ramdisk

write througput:
(cp -a)
		small files		big files
test1		605 MB/s		722 MB/s
test2		1006-1266 MB/s		1327-1724 MB/s
test3		568 MB/s		654 MB/s
test4		1356 MB/s (466 MB/s)	1647 MB/s (514 MB/s)
test5		1471 MB/s (375 MB/s)	1851 MB/s (406 MB/s)
test6		1899 MB/s		2236 MB/s
test7		1277 MB/s (806 MB/s)	1446 MB/s (798 MB/s)

(the number in parentheses is the throughput including the 'sync' command)


read throughput:
(grep -Fr blabla /mnt/test)
		small files		big files
test1		566 MB/s		1486 MB/s
test2		563 MB/s		1471 MB/s
test3		548 MB/s		1393 MB/s
test4		512 MB/s		1293 MB/s
test5		516 MB/s		1275 MB/s
test6		664 MB/s		2386 MB/s
test7		561 MB/s		1495 MB/s

read from cache:
		small files		big files
		669 MB/s		2423 MB/s


linear writes:
(dd if=/dev/zero of=/mnt/test/file bs=64k count=10000)
		async		oflag=dsync 
test1		744 MB/s	737 MB/s
test2		1.4 GB/s	1.4 GB.s
test3		456 MB/s	288 MB/s
test4		2.0 GB/s	260 MB/s
test5		2.0 GB/s	151 MB/s
test6		2.9 GB/s	2.9 GB/s
test7		2.0 GB/s	501 MB/s


compile time benchmark:
- a test, how well the compiler performs if the executable "cc1" runs
  directly from persistent memory
(make -j1 in the lvm source tree)
without dax:		33.6s
with dax on pmem:	35.0s	(4% worse)


dm-writecache throughput:
(dd if=/dev/zero of=/dev/mapper/wc bs=64k oflag=direct)
writecache block size	512		1024		2048		4096
movnti			496 MB/s	642 MB/s	725 MB/s	744 MB/s
clflushopt		373 MB/s	688 MB/s	1.1 GB/s	1.2 GB/s
with patch		497 MB/s	688 MB/s	1.1 GB/s	1.2 GB/s