Speed test of SD card

[Fraoch]

Yeah, I know, this isn't possible, but it's like I'm reading from a cache or something?

Doing speed testing on an SD card using Ubuntu 12.10 64-bit and the following commands:

dd if=/dev/zero of=~/test.tmp bs=500K count=1024

dd if=~/test.tmp of=/dev/null bs=500K count=1024

I'm interpreting that as writing zeroes to test.tmp, then reading back test.tmp to /dev/null. Right?

It worked a couple of times, now I'm getting:

mark@Sauron:/media/mark$ dd if=/dev/zero of=/media/mark/test.tmp bs=500K count=1024
1024+0 records in
1024+0 records out
524288000 bytes (524 MB) copied, 466.175 s, 1.1 MB/s
mark@Sauron:/media/mark$ dd if=/media/mark/test.tmp of=/dev/null bs=500K count=1024
1024+0 records in
1024+0 records out
524288000 bytes (524 MB) copied, 0.166738 s, 3.1 GB/s

What? 3.1 GB/s shouldn't be possible over USB 2.0. It shouldn't even be possible over SATA3 for that matter...

What's going on here?

[The Cog]

The second dd is probably reading from cache memory rather than from the hardware.

[Fraoch]

The second dd is probably reading from cache memory rather than from the hardware.

That's what it looks like, but I can't figure out what cache. The SD card itself has no cache of course, nor does the card reader. With a speed of 3.1 GB/s, looks like it's reading from main memory somehow? Odd also that it performed this test properly a few times, then this, consistently and repeatedly.

[rnerwein]

That's what it looks like, but I can't figure out what cache. The SD card itself has no cache of course, nor does the card reader. With a speed of 3.1 GB/s, looks like it's reading from main memory somehow? Odd also that it performed this test properly a few times, then this, consistently and repeatedly.

hi
if you want to know what's going up - following these steps.
1. after you reboot - open a terminal and run: top (have a look at mem used and cached used)
2. run your tests using another filename ---> look again at the values
3. run it again ---> look at your values
4. delete those files
5. look again at the top values (mem used and cache) - surprised ?-)
what i'll tell U is - it is not always what you see is what you get.
by the way that's a part of the reason (I/O speed - not CPU ticks - that's mostely blended ) of the price of hig-end servers.
ciao

[Cheesemill]

That's what it looks like, but I can't figure out what cache. The SD card itself has no cache of course, nor does the card reader. With a speed of 3.1 GB/s, looks like it's reading from main memory somehow? Odd also that it performed this test properly a few times, then this, consistently and repeatedly.

It's being cached in RAM.

Any file that you access on a Linux machine is held in RAM as a cached file, the file is only dropped by the cache if your RAM is needed for a running process.
For more information this page may be of some help...

http://www.linuxatemyram.com/

[rnerwein]

It's being cached in RAM.

Any file that you access on a Linux machine is held in RAM as a cached file, the file is only dropped by the cache if your RAM is needed for a running process.
For more information this page may be of some help...

http://www.linuxatemyram.com/

+1 Cheesemill - short and clear (just a little question: what is RAM ? - you now what i mean - have my funny day today )
ups - don't understand me false ( i think you now what i mean)
ciao
richi

[Fraoch]

It's being cached in RAM.

Any file that you access on a Linux machine is held in RAM as a cached file, the file is only dropped by the cache if your RAM is needed for a running process.
For more information this page may be of some help...

http://www.linuxatemyram.com/

That makes sense, thank you.

I'm googling around, and it seems that there are workarounds to disable caching for this process:

http://romanrm.ru/en/dd-benchmark

I might try the conv=fdatasync option specified on that page.

Incidentally I probably wasn't using this the proper way in the first place. I'm speed testing two new SD cards for use in a Raspberry Pi (not delivered yet!) which only has 512 MB of RAM and could never cache 524 MB of data:

http://elinux.org/RPi_SD_cards#Performance

[Cheesemill]

Just a little snippet to add. If you want to force all cached files to be dropped to ensure you are testing the speed of the SD card you can use the following command between runs...

echo 3 | sudo tee /proc/sys/vm/drop_caches

[tgalati4]

Because of the hardware differences between a Rapberry Pi and a true Linux desktop or laptop, the real speed test will need to be done using actual hardware, running actual programs on the Pi. Anything else is just a suggestion of performance. If you want robust testing, then install phoronix-test-suite and run some I/O and file-handling benchmarks. I don't know if the phoronix test suite has been ported to the Pi, but that would give you some serious benchmarks to compare.

Because of the wear-leveling used on SD cards, they tend to slow down over time. So new card versus 6-month-old card testing would be useful.

[rnerwein]

Just a little snippet to add. If you want to force all cached files to be dropped to ensure you are testing the speed of the SD card you can use the following command between runs...

echo 3 | sudo tee /proc/sys/vm/drop_caches

hi cheesemill
that's ok - but i think you missed the "sync" before. must be your keyboard
no swet - have fun (you understand me or ?)
ciao