[550D] Card benchmarks

[Alex]

Here's a way to do a simple benchmark for card read/write speed: with
this build, select Debug->Card benchmark.

It will write a 256MB file (RAM dump), and then it will read it. The
test is done with 3 buffer sizes.

After a few minutes of running, it will display the results.

Before running the test, set the Auto power off to a few minutes (I
have it on 8 minutes). You don't want the camera to turn off in the
middle of the test.

The results may also depend on:
- Camera mode / video resolution, or maybe just if it's in LiveView or not
- Card fragmentation (in theory, you get best results with a formatted card)
- ML settings (especially those in LiveV); it's better to do this test
without any config file.
- Whether you press buttons or not during the test :)

This build also contains a few UI tweaks for zoom overlay (regarding
zebras and false colors). See the docs for details.

[Alex]

Some numbers from dvxuser forum:
----------------------------------
kgimedia:
FIRST CARD CHECK

Adata Class 10 16 GB
WS= 16k 2.6 MB/s
RS= 16k 3.6 MB/s
WS= 128k 9.4 MB/s
RS= 128k 13.4 MB/s
WS= 16384k 17.0 MB/s

Micro Center Class 10 8 GB
WS= 16k 3.2 MB/s
RS= 16k 4.0 MB/s
WS= 128k 9.8 MB/s
RS= 128k 14.2 MB/s
WS= 16384k 14.2 MB/s
----------------------------------
bluesgeek:
Transcend TS16GSDHC10

WS @16k = 2.6 MB/s
RS @16k = 3.2 MB/s

WS @128k = 11.6 MB/s
RS @128k = 12.1 MB/s

WS @16384k = 19.6

I can't use QScale dependable on these cards over -5.
----------------------------------

It seems the overhead of a FIO_WriteFile is significant. Not sure what
is the buffer size used by Canon when recording. AJ, do you have any
idea?

[Antony Newman]

Alex,

I haven't experimented in this area (other than noting that some C implemented functions like fwrite are very slow).

The terminaldriver that writes movies may run at different speed from the read/write in your benchmark.

Is the following function a 'lower level' file manager subroutine (I'm not sure)?    fmExecCreateFile()


I think to get the highest writing rate we would need to:
+) Dump out the %CPU of the DryOs tasks.
+) Change the priority of the tasks that asynchronously writes to have a higher priority
+) Change the priority of the are not involved to lower priority.


AJ.

[Piers]

Dunno if I should post this on dvxuser but ...

Kingston 16G class 4

16k:

2.9w 3.9r

128k

8.8w, 14.2r

16384

11.6w, read skipped (huh?)

[chungdha]

- Toshiba 16gb class 4 -

16k: 2.8w, 3.9r

128k: 7.5w, 14.2r

16384: 9.4w
Read test skipped

- Toshiba 16gb class 6 -

16k: 3.5w,4r

128k: 12.8w,14.2r

16284: 18.2w
read test skipped.

Seem read test skipped is standard thing on the end.

[Pash0k]

RIDATA 16gb 10 class

WS= 16k 3.0 MB/s
RS= 16k 4.0 MB/s
WS= 128k 12.8 MB/s
RS= 128k 14.2 MB/s
WS= 16384k 18.2 MB/s

[Pash0k]

Transcend 16gb 10 class

WS= 16k 3.0 MB/s
RS= 16k 4.0 MB/s

WS= 128k 9.1 MB/s
RS= 128k 13.4 MB/s
WS= 16384k 12.1 MB/s

[Piers]

Delkin 16G Class 6 ("22MB/s 150x" written on it) - bought it from B&H

      W     R

16k   3.1   4.3

128k  9.1  14.2

16M  17.0   x

[chungdha]

Notice this build the audio does not jump to External directly on
start up even record it would record with internal even when the
setting is external. Need to hit playback first before it would record
with the external mic.

>  magiclantern-2011Mar14.550d.fw109.AudioMon.card-benchmark.alex.zip
> 1276KViewDownload

[Piers]

I've noticed funkiness with the auto input, but have nothing repeatable to report yet.

Working hypothesis: does not set audio on startup (even though menu/meters displays "ext", it clearly isn't) but then, having cycled through inputs back to auto, it does use the ext input - even after a power cycle.

Testing continues ....