I have done some testing of the various SD cards I have in my
collection. I used the C program, Bonnie, much used in the past. It
is relatively small and uncomplicated, at least relatively so:) It
compiles fine on the Ben with no problems with the latest image. I
used whatever the default values were for compiling. Perhaps some of
these numbers would differ if everything in Bonnie were tuned to the
precise CPU in the Ben. That is over my head and in the past I found
conflicting advice in a web search (That's nothing new, either!).
Maybe some C and Mips expert could spell out what gcc options might be
best:)
Outline of procedure:
1. Bonnie was run from the rootfs in the Nand partition.
2. Each card had two partitions on it: one that was about 1 GB in
size and the other whatever was left over. The 1 GB partition was
called "small" (S) and the other was call "big" (B) file system.
3. Three different filesystems were tested: ext2, ext4, and in a more
limited sample, ext4 without a journal.
4. Each file system was empty so the only file present during the
tests was the one written by Bonnie.
5. The mount options were always: noatime,nodiratime.
6. Eight runs of Bonnie were made for each of the two different sizes
of files systems. Thus there were 16 runs for each file system. Each
run wrote and read 128 MB more than once. The 16 runs took
approximately one hour on the Ben.
8. The average value of the eight runs is reported in the table
below. I also computed the standard deviation, the minimum, and the
maximum. These are not reported to save space.
9. What I call the part number is what I thought the part number was.
On some cards there are as many as three sets of numbers--I don't know
which ones are important and which ones are not. The Transcend part
number that was too long is: MMAGR08GUDCA-DB and the one for the Cat
10 Patriot is: N0303-001.A00LF
10. The speeds in the table are given in Kbytes/second as reported by
the default configuration for Bonnie.
-----------------------------------------------------------------------------
Summary of Bonnie I/O Tests on
Various Micro-SD Cards.
Make Part# Cap Cat FlSystm FsSz CW BCW BCRW CR BCR
------ -------- --- --- ------- ---- ---- ---- ---- ---- ----
Kngstn K33045Y 8 4 Ext2 B 2378 2348 1859 3703 7989
S 2437 2307 1864 3715 7979
Ptrt E08213G 16 4 Ext2 B 3314 4053 3288 3871 7930
S 3373 4017 3234 3871 7977
Kngstn K33045Y 8 4 Ext4 B 3462 4766 2952 3871 7784
S 3464 4790 2971 3850 7855
Ptrt E08213G 16 4 Ext4 B 3431 4780 2950 3844 7977
S 3465 4821 2954 3865 8002
Make Part# Cap Cat FlSystm FsSz CW BCW BCRW CR BCR
------ -------- --- --- ------- ---- ---- ---- ---- ---- ----
Trnscnda 9153BA 8 6 Ext2 B 3736 5651 3510 3918 7965
S 3416 4607 3258 3885 7963
8 6 Ext4 B 3732 5510 3506 3920 8002
S 3690 5289 3410 3904 7973
8 6 Ext4nj B 3814 5850 3619 3918 7947
S 3826 5930 3639 3918 7985
Trnscndb MMA... 8 6 Ext2 B 3601 5173 3286 3825 7884
S 3609 5205 3299 3822 7900
Ext4 B 3632 5285 3348 3891 7878
S 3646 5247 3325 3900 7883
Ext4nj B 3755 5625 3415 3880 7873
S 3789 5638 3433 3887 7881
Trnscndc MMA... 8 6 Ext2 B 3582 5171 3248 3807 7854
S 3610 5089 3264 3817 7837
Ext4 B 3622 5241 3312 3884 7763
S 3601 5172 3265 3881 7842
Trnscndd 9153BA 8 6 Ext2 B 3727 5641 3511 3926 7973
S 3449 4578 3249 3887 7989
Ext4 B 3732 5549 3538 3915 7984
S 3669 5278 3410 3904 7936
Trnscnde MMA... 8 6 Ext2 B 3644 5164 3295 3830 7872
S 3653 5268 3351 3827 7887
Ext4 B 3651 5290 3366 3899 7855
S 3551 5265 3356 3898 7859
Make Part# Cap Cat FlSystm FsSz CW BCW BCRW CR BCR
------ -------- --- --- ------- ---- ---- ---- ---- ---- ----
Kngstn N0302.... 8 10 Ext2 B 3723 5488 3106 3874 7634
S 3719 5266 3033 3876 7642
Ext4 B 3285 5198 2988 3871 7647
S 3269 4434 2648 3816 7626
Ext4nj B 3689 5555 3160 3882 7645
S 3701 5269 2984 3878 7624
-----------------------------------------------------------------------------
Read or Write rate in KBytes per second.
Cat- Category
Cap - capacity in GB
CW - character by character write
BCW - write a block of characters
BCRW - change a character and re-write the block
CR - character by character read
BCR - Read characters by block
Ext4nj - Ext4 with no journal
----------------------------------------------------------------------------
Observations on these results:
1. The difference between the speeds for Ext2 and Ext4 depends on the
card, and in some cases on the size of the file system. So reasons
other than write speed probably are more important, unless your card
is one that is sensitive to the difference. The effect of wear from a
journal is discussed on the Web with conclusions that vary:) I
currently use Ext4 with a journal for my big data partition on a 32GB
Transcend Cat 4 card and ext2 on the root partition (I believe this is
required to boot because when I tried Ext4 I could not boot from the
card.)
2. Read speeds are about the same across the cards. Perhaps the rate
is limited by the speed of the CPU or the bus used in the Ben.
3. Transcend cards that have the same category and capacity but
different part numbers do seem to differ a bit in their writing speed.
This difference is larger than the variability in each cards speeds so
it looks real.
4. The Cat 10 card proves to be a bit disappointing in that is about
the same as the better Cat 6 cards. Again, this might be a limit set
by the Ben and not the card. No way to tell unless the card is tested
on a more powerful machine with limited RAM. The size of the file
written should be a good bit larger than installed RAM to minimize the
effect of data caching.
5. Using Ext4 without a journal is slightly faster on writing.
Looking at these numbers a Cat 6 card looks like the best one for the
Ben. However, when I searched for 32GB cards, I found only Cat 4 and
Cat 10 cards.
This was interesting and time consuming. Probably won't do it again:)
Delbert
_______________________________________________
Qi Hardware Discussion List
Mail to list (members only): [email protected]
Subscribe or Unsubscribe:
http://lists.en.qi-hardware.com/mailman/listinfo/discussion
