On Wednesday 21 September 2011, David Yip wrote:
Ok, here is the result:
Cc'd the mailing list.
Thanks for the update, I'm adding this to the wiki now.
Arnd
vpcf1 ~ # head /sys/block/mmcblk0/device/{cid,csd,scr,*rev,*id,name,serial}
==> /sys/block/mmcblk0/device/cid <== 744a45534443202000000004c000b700
==> /sys/block/mmcblk0/device/csd <== 400e00325b59000077957f800a400000
==> /sys/block/mmcblk0/device/scr <== 0235800000000000
==> /sys/block/mmcblk0/device/fwrev <== 0x0
==> /sys/block/mmcblk0/device/hwrev <== 0x0
==> /sys/block/mmcblk0/device/cid <== 744a45534443202000000004c000b700
==> /sys/block/mmcblk0/device/manfid <== 0x000074
==> /sys/block/mmcblk0/device/oemid <== 0x4a45
==> /sys/block/mmcblk0/device/name <== SDC
==> /sys/block/mmcblk0/device/serial <== 0x000004c0
vpcf1 ~ # fdisk -lu /dev/mmcblk0
Disk /dev/mmcblk0: 16.1 GB, 16050552832 bytes 255 heads, 63 sectors/track, 1951 cylinders, total 31348736 sectors Units = sectors of 1 * 512 = 512 bytes Sector size (logical/physical): 512 bytes / 512 bytes I/O size (minimum/optimal): 512 bytes / 512 bytes Disk identifier: 0x000d6d26
Device Boot Start End Blocks Id System/dev/mmcblk0p1 8192 31342591 15667200 83 Linux
vpcf1 flashbench # ./flashbench --open-au --open-au-nr=5 --random --blocksize=512 --erasesize=$[4 * 1024 * 1024] /dev/mmcblk0 4MiB 7.45M/s 2MiB 6.61M/s 1MiB 7.59M/s 512KiB 7.63M/s 256KiB 7.52M/s 128KiB 7.43M/s 64KiB 7.35M/s 32KiB 5.56M/s 16KiB 3.68M/s 8KiB 1.49M/s 4KiB 750K/s 2KiB 366K/s 1KiB 185K/s 512B 93.3K/s vpcf1 flashbench # ./flashbench --open-au --open-au-nr=5 --random --blocksize=1024 --erasesize=$[4 * 1024 * 1024] /dev/mmcblk0 4MiB 8.36M/s 2MiB 8.34M/s 1MiB 8.3M/s 512KiB 8.27M/s 256KiB 8.22M/s 128KiB 8.07M/s 64KiB 7.89M/s 32KiB 5.47M/s 16KiB 3.56M/s 8KiB 1.46M/s 4KiB 739K/s 2KiB 361K/s 1KiB 176K/s vpcf1 flashbench # ./flashbench --open-au --open-au-nr=5 --random --blocksize=2048 --erasesize=$[4 * 1024 * 1024] /dev/mmcblk0 4MiB 7.07M/s 2MiB 7.02M/s 1MiB 6.99M/s 512KiB 6.95M/s 256KiB 6.94M/s 128KiB 6.83M/s 64KiB 6.67M/s 32KiB 5.52M/s 16KiB 3.66M/s 8KiB 1.48M/s 4KiB 734K/s 2KiB 367K/s vpcf1 flashbench # ./flashbench --open-au --open-au-nr=5 --random --blocksize=4096 --erasesize=$[4 * 1024 * 1024] /dev/mmcblk0 4MiB 7.36M/s 2MiB 6.96M/s 1MiB 6.99M/s 512KiB 6.95M/s 256KiB 6.92M/s 128KiB 6.82M/s 64KiB 6.67M/s 32KiB 5.49M/s 16KiB 3.66M/s 8KiB 1.48M/s 4KiB 733K/s vpcf1 flashbench # ./flashbench --open-au --open-au-nr=5 --random --blocksize=8192 --erasesize=$[4 * 1024 * 1024] /dev/mmcblk0 4MiB 4.05M/s 2MiB 7.05M/s 1MiB 6.98M/s 512KiB 6.96M/s 256KiB 6.91M/s 128KiB 6.76M/s 64KiB 6.59M/s 32KiB 5.42M/s 16KiB 2.63M/s 8KiB 1.46M/s
Hope they are what you need : )
It's enough, yes. The blocksize argument actually doesn't really affect anything, except how many tests are run -- with blocksize=8192 it just stops earlier than with blocksize=1024.
What I normally need is the result of one run with --nr-open-au just after the threshold, i.e 6 in this case. Since you already had those before, it's fine.
On Wed, Sep 21, 2011 at 2:45 AM, Arnd Bergmann arnd.bergmann@linaro.orgwrote:
On Tuesday 20 September 2011, David Yip wrote:
Hi Arnd,
I did few tests of Transcend SDHC CL6 16GB SD Flash.
http://www.picstop.co.uk/media/catalog/product/cache/1/image/5e06319eda06f02...
The erase size block is 4MB, I *believe*.
Hi David,
Yes, this card is pretty much standard, 4MB erase blocks, 5 open simultaneously. I've added it to the wiki, but it would be nice if you could provide the remaining data, similar to the commands I posted in http://lists.linaro.org/pipermail/flashbench-results/2011-March/000039.html
For a simple card like this, doing a "flashbench -a --blocksize=1024" test should show the erase block size very clearly, so you don't need to do all the tests. I would also strongly recommend not scripting around flashbench but doing it one run at a time, which is much faster (you can kill it as soon as you see that it's clearly within the pathological path of the --open-au test) and does not wear out the card that much. For the number of open erase blocks, the only thing that matters really is the highest number that is fast, followed by one after that which is slow.
What I want to add to the wiki along with this is
- complete identification (from /sys/block/mmcblk0/device/* on SD/mmc
or hdparm -I on CF/ATA)
- exact size (from /proc/partitions or fdisk -lu)
- page size (do one fast test run with --blocksize=512 to try all
block size)
I'd definitely like to see more results, especially if you have other SSDs or USB drives. It would also be nice to post the results inline in the mail instead of an attachment and to take flashbench-results@lists.linaro.org on Cc: so they end up in the online archive.
Thanks,
Arnd