On Sat, 06 Oct 2012 15:13:32 +0200 "Holger Bast" HolgerBast@gmx.de wrote:

I've tested another SDHC card. But this time it was a class-4 mini-SDHC with 8GB. The card is a very cheap one from the German drugstore DM. (No, usually I do not buy my electronics in drugstores ;-) - just a backup card.)

Can you provide brand info, part number, and class rating?

Ok, here are the results. I'm not sure, if my interpretation of them is right.

I've read, that the choice of the underlaying filesystem depends on the amount of open erase blocks. Is there any rule of best practice for choosing the right filesystem for a particular sd card?

btrfs seems to perform best if you have 3 or more open erase block ability. ext4 isn't quite as good but still much better than ext3.

If you only have 1 open erase block, FAT is really the only choice I know of to get decent performance but then you trade off quite a lot in normal *nix operation, like permissions on files.

$ fdisk -l /dev/sdc

Disk /dev/sdc: 7948 MB, 7948206080 bytes 255 heads, 63 sectors/track, 966 cylinders, total 15523840 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: 0x00000000

Can you try this card in an sd/mmc reader that shows up as /dev/mmcblk device in Linux? If so, can you run

$ head /sys/block/mmcblk0/device/* 2>/dev/null | grep -v ^$

$ ./flashbench -a /dev/sdc --blocksize=1024 align 2147483648        pre 825µs       on 1.19ms       post 851µs    diff 347µs align 1073741824        pre 841µs       on 1.18ms post 830µs      diff 341µs align 536870912         pre 821µs       on 1.14ms       post 779µs      diff 341µs align 268435456         pre 842µs       on 1.18ms       post 818µs      diff 346µs align 134217728         pre 832µs       on 1.16ms       post 799µs  diff 346µs align 67108864          pre 860µs       on 1.16ms post 822µs      diff 321µs align 33554432          pre 832µs       on 1.15ms       post 802µs      diff 331µs align 16777216          pre 783µs       on 1.16ms       post 824µs      diff 359µs align 8388608           pre 807µs       on 1.1ms        post 781µs      diff 303µs align 4194304           pre 817µs       on 1.14ms       post 786µs    diff 340µs align 2097152           pre 847µs       on 924µs  post 791µs      diff 105µs align 1048576           pre 791µs on 935µs        post 807µs      diff 136µs align 524288  pre 826µs       on 906µs        post 823µs      diff 81.9µs align 262144            pre 810µs       on 926µs        post 826µs  diff 108µs align 131072            pre 832µs       on 913µs  post 801µs      diff 96.5µs align 65536             pre 819µs on 922µs        post 830µs      diff 97.2µs align 32768 pre 831µs       on 941µs        post 807µs      diff 122µs align 16384             pre 772µs       on 918µs        post 820µs  diff 122µs align 8192              pre 798µs       on 845µs  post 837µs      diff 27µs align 4096              pre 794µs       on 823µs        post 806µs      diff 22.2µs align 2048              pre 836µs       on 850µs        post 801µs      diff 31.7µs

## Looks like 4MB erase blocksize and 16KB page size??

Appears so.

$ ./flashbench /dev/sdc --open-au --erasesize=$[4*1024*1024] --blocksize=$[4*1024] --random --open-au-nr=1 4MiB    5.41M/s 2MiB    8.97M/s 1MiB    5.98M/s 512KiB  17.1M/s 256KiB  10.9M/s 128KiB  4.25M/s 64KiB   2.13M/s 32KiB   1.06M/s 16KiB   545K/s   8KiB    2.07M/s 4KiB    1.76M/s

$ ./flashbench /dev/sdc --open-au --erasesize=$[4*1024*1024] --blocksize=$[4*1024] --random --open-au-nr=2 4MiB    8.21M/s 2MiB    5.75M/s 1MiB    18.5M/s 512KiB  16.8M/s 256KiB  10.7M/s 128KiB  4.26M/s 64KiB   2.11M/s 32KiB   1.06M/s 16KiB   545K/s   8KiB    2.05M/s 4KiB    1.69M/s

$ ./flashbench /dev/sdc --open-au --erasesize=$[4*1024*1024] --blocksize=$[4*1024] --random --open-au-nr=3 4MiB    9.18M/s 2MiB    19.1M/s 1MiB    18.7M/s 512KiB  16.5M/s 256KiB  10.8M/s 128KiB  4.23M/s 64KiB   2.11M/s 32KiB   1.06M/s 16KiB   544K/s   8KiB    1.99M/s 4KiB    1.67M/s

$./flashbench /dev/sdc --open-au --erasesize=$[4*1024*1024] --blocksize=$[4*1024] --random --open-au-nr=4 4MiB    10.6M/s 2MiB    19.3M/s 1MiB    18.6M/s 512KiB  16.6M/s 256KiB  10.9M/s 128KiB  4.24M/s 64KiB   2.11M/s 32KiB   1.06M/s 16KiB   544K/s   8KiB    1.86M/s 4KiB    1.53M/s

$ ./flashbench /dev/sdc --open-au --erasesize=$[4*1024*1024] --blocksize=$[4*1024] --random --open-au-nr=5 4MiB    6.1M/s   2MiB    6.68M/s 1MiB    4.39M/s 512KiB  2.96M/s 256KiB  1.95M/s 128KiB  1.02M/s 64KiB   570K/s   32KiB   334K/s   16KiB   167K/s   8KiB    1M/s     4KiB    1.14M/s

$ ./flashbench /dev/sdc --open-au --erasesize=$[4*1024*1024] --blocksize=$[4*1024] --random --open-au-nr=6 4MiB    7.5M/s   2MiB    4.69M/s 1MiB    3.38M/s 512KiB  2.02M/s 256KiB  1.22M/s 128KiB  688K/s   64KiB   406K/s   ^C

## Looks like 4.

Yes, for random.

Can you run the --open-au tests without --random?

$./flashbench /dev/sdc -f --erasesize=$[4*1024*1024] 4MiB    4.11M/s  17.4M/s  16.9M/s  5.62M/s  5.18M/s  5.2M/s   2MiB    5.02M/s  17.6M/s  17.2M/s  5.32M/s  5.03M/s  5.21M/s   1MiB    5.28M/s  17.2M/s  17.2M/s  5.53M/s  5.21M/s  5M/s     512KiB  5.04M/s  17.4M/s  5.07M/s  5.21M/s  5.27M/s  5.01M/s   256KiB  5.19M/s  17.1M/s  5.02M/s  5.21M/s  5.02M/s  5.21M/s   128KiB  2.35M/s  16.6M/s  4.96M/s  5.22M/s  4.99M/s  5.18M/s   64KiB   4.98M/s  16.6M/s  5.21M/s  4.91M/s  4.97M/s  5.21M/s   32KiB   2.31M/s  14.8M/s  5.11M/s  4.92M/s  5.12M/s  5.2M/s   16KiB   4.81M/s  12.4M/s  5.07M/s  5M/s     5M/s     4.82M/s

## Looks like the second and the third block were faster than the others.

Looks like just the 2nd is faster through the entire test, but the 3rd is faster only at larger sizes.

-Andrew