On Wednesday 13 July 2011, Peter Warasin wrote:
hi guys
this card seems quite unqualified
Agreed
==> /sys/block/mmcblk0/device/manfid <== 0x000002
==> /sys/block/mmcblk0/device/oemid <== 0x544d
"TM" -- Toshiba Memory (?), same as Kingston cards
$ ./flashbench /dev/sda3 --open-au-nr=1 -O --blocksize=2048 4MiB 6.62M/s 2MiB 4.95M/s 1MiB 4.99M/s 512KiB 4.94M/s 256KiB 4.88M/s 128KiB 4.64M/s 64KiB 6M/s 32KiB 4.61M/s 16KiB 3.91M/s 8KiB 3.15M/s 4KiB 2.4M/s 2KiB 25.3K/s
$ ./flashbench /dev/sda3 --open-au-nr=2 -O --blocksize=2048 4MiB 5.68M/s 2MiB 3.05M/s 1MiB 1.56M/s 512KiB 797K/s 256KiB 404K/s 128KiB 205K/s 64KiB 103K/s 32KiB 51.6K/s freezing
-> 1 block
Right, also like Kingston behaviour. This card also has the extremely annoying property of not being able to do efficient <4KB writes at all. It has a page size of 4KB, and if you have partitions that are not aligned to 4KB, you end up in the 25K/s category instead of the 2.4M/s+ category even for simple linear writes.
$ ./flashbench /dev/sda3 --open-au-nr=1 -O --blocksize=2048 --random 4MiB 3.33M/s 2MiB 3.3M/s 1MiB 3M/s 512KiB 1.2M/s 256KiB 671K/s 128KiB 310K/s 64KiB 150K/s 32KiB 81.9K/s 16KiB 40.5K/s 8KiB 20.6K/s ^C
This too.
-> unqualified for random access
$ ./flashbench /dev/sda3 -f --erasesize=$[4*1024*1024] 4MiB 3.31M/s 5.06M/s 5.02M/s 5.01M/s 4.93M/s 5M/s 2MiB 4.96M/s 5.04M/s 4.98M/s 5.01M/s 5.02M/s 5.04M/s 1MiB 4.99M/s 5.01M/s 4.98M/s 5.01M/s 5.01M/s 5.02M/s 512KiB 4.99M/s 5.06M/s 5.04M/s 4.96M/s 4.97M/s 4.95M/s 256KiB 4.85M/s 5.01M/s 4.98M/s 5M/s 5.03M/s 5.05M/s 128KiB 4.66M/s 4.84M/s 4.87M/s 4.88M/s 4.93M/s 4.88M/s 64KiB 6M/s 5.47M/s 5.43M/s 5.11M/s 5.35M/s 5.38M/s 32KiB 4.59M/s 5.27M/s 5.23M/s 5.29M/s 5.2M/s 5.21M/s 16KiB 3.87M/s 4.29M/s 4.3M/s 4.24M/s 4.29M/s 4.3M/s
what can i say with these?
What this test does is do the same test as '--open-au-nr=1 -O' on the first six erase blocks (default for --fat-nr=X), to see if any of them are faster or slower than others. On some cards, the differences are quite noticeable.
If you also pass --random, you should notice that some of the columns actually are not too bad and stay reasonably fast, while the other columns show the same results as the --open-au --random test above.
Probably the first two columns are fast, while the others are slow, which is consistent with regular Kingston cards.
Thanks for the data point,
Arnd