Goal

Get a good comparison between various SD Cards of various sizes, brands, and classes.

Determine if write bottlenecks are due to the Overo's data bus, or the microSDHC.

Results

Write and Read speeds are with booting from NAND.

TODO try jfs and ext4

Manufacturer      Size      Class     Write Speed     Read Speed    Filesystem
------------      ----      -----     -----------     ----------    ----------

KingMax           8GB       10        4.7 MB/s        14.1 MB/s     ext3,defaults

Kingston          16GB      10        4.9 MB/s        13.2 MB/s     ext3,defaults

Kingston          16GB      10        2.8 MB/s        13.5 MB/s     **ext2**,defaults

Kingston          16GB      4         4.9 MB/s        12.6 MB/s     ext3,defaults

SanDisk           16GB      2         3.1 MB/s        13.1 MB/s     ext3,defaults

? (Gumstix)       2GB       ?         3.2 MB/s        9.7 MB/s      ext3,defaults

Comparison: regular desktop with Kingston Multi-reader USB 2.0

Kingston          16GB      10        5.3 MB/s        137.0 MB/s    **ext4**,defaults

Kingston          16GB      10        3.3 MB/s        36.0 MB/s     ext3,defaults

Kingston          16GB      10        5.5 MB/s        91.2 MB/s     **ext2**,defaults

Procedure

I would like to try both configurations of booting from the NAND and booting from the microSDHC.

NOTE: The results above are only from booting the NAND.

Booting from the NAND

Connect via Serial:
1. ls /dev/tty*[uU][sS][bB]* # Should list usb-serial devices equally well on Mac and Linux
2. sudo minicom -s # Configure to use the port listed above
Power on and enter u-boot by pressing the any key
run nandboot

If the card wasn't already formatted, it's formatted as ext3

mount | grep mmcblk0 | grep -v ext3
umount /media/mmcblk0p1/
mkfs.ext3 /dev/mmcblk0p1
mount /dev/mmcblk0p1 /media/mmcblk0p1

TODO: should have formatted all cards with filesystem before testing

Booting from the microSDHC

There are two other posts (possibly in the related posts section) about partitioning and formatting the microSDHC. Once that's done, they'll boot.

Benchmark Procedure

What we need to know is if the card can stream the MB it boasts or not.

NOTE: bitbake bonnie++ provides a good benchmarking tool for more realistic testing.

# Get some random bytes from memory - say 40MB or so
sync
dd if=/dev/urandom of=/dev/shm/rand.dd bs=1M count=40

# Emulate a continuous stream of data
sync
MMC=/media/mmcblk0p1
#NOTE: This fills the cache buffers with data 
cat /dev/shm/rand.dd /dev/shm/rand.dd | dd of=${MMC}/rand.dd

# The overhead of the un`sync`d data above will make this more accurate
cat /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd \
  /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd \
  /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd \
  /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd \
  /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd \
  /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd \
  /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd \
  /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd \
  /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd \
  /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd | dd of=${MMC}/rand.dd

# Test read speed just for fun
sync
dd if=${MMC}/rand.dd of=/dev/shm/rand.dd

NOTE: By default these devices are mounted with the sync option. You'll get a whopping 6.4KB/s unless you remount them.

NOTE: Due to caching, the speeds appear slightly faster than they really are. If you just test with a small amount of data - 40MB, for example - you'll see that quite profoundly.

Appendix

Generating Random Bits

sync
dd if=/dev/urandom of=/dev/shm/rand.dd bs=1M count=40
41943040 bytes (42 MB) copied, 42.311 s, 991 kB/s

Writing to the NAND (which only had 45 MB free)

45474304 bytes (45 MB) copied, 414.42 s, 110 kB/s

Updated at 2010-09-01

[Home] testing file write speeds