next »

[Carol]

Does the rated speed of a SD card matter when it's used in the XO's SD card slot? 

[Carol]

Hello.  Anybody out there?

[olpc_franco]

Yes, franco is here but unqualified to answer the question.  Patience.

[eden]

In my limited experience, SD card speed seems to matter most if you are using a digital camera.  Newer digital cameras tend to require the faster cards so they can record video smoothly and also write the larger photo files.

I have a class 2 4GB Lexar SDHC card with some videos on it that I have watched on the XO with mplayer.  Any lack of quality in the playback is probably from the way I encoded the video.  I haven't noticed any lagging.  I have also written data to the card and it seems fast enough to me.  Before I splurged for the newer card, I was using an ancient 64MB SD card.  Reading and writing to that also seemed to go at an okay speed, maybe a hair slower than I get now. 

I searched the OLPC Wiki and found their page on SD cards: http://wiki.laptop.org/go/SD
It doesn't yet mention speed.

[akpoff]

Does the rated speed of a SD card matter when it's used in the XO's SD card slot? 

Hi Carol,
As Eden pointed out, OLPC haven't released any specs on the SD card slot's speed rating.  However, we might reasonably infer that it could benefit from faster SD cards by noting that the XO can use SDHC (high capacity).  It's not a sure fire way of knowing but it's a clue.

When I get my XO I'll try some speed tests with various speed cards I have on hand unless someone beats me to it.

One thought, if you watch the sales you can get class 4 and 6 cards for just a bit more than the slower cards.  Might be worth the extra couple of bucks.

--Aaron

[cmonkey]

hdparm -t on my 8gb A-Data SDHC card gives 1.86 MB/s buffered read.

On an old 512mb Sandisk regular SD card, I get 1.20 MB/s.

hdparm isn't the greatest tool to benchmark a drive with, especially not when it's flash, but it does show a difference between the two.

[ossguy]

I performed some speed tests on the XO as well as an Eee PC and a generic USB SD reader.  The main thread for the announcement can be found at http://olpcnews.com/forum/index.php?topic=2145.0, which links to my blog post about it.

Overall, the results were somewhat disappointing.  The XO's SD card reader lags far behind the Eee PC and the USB reader.

I would be interested in hearing reasons for the poor performance; it is a bit strange considering that the SD card reader is built right onto the XO's logic board.

I haven't found the slow performance of the XO's SD reader to be an issue; most of the things I do with it don't require fast read/write speeds.

Denver

[LaPaglia]

I'm sure if I follow the links I could find the answer but humor me please. Which version of the OS are you using on the XO you used?

[ossguy]

Yes, you could find it by following the links, but it's buried at the bottom of the blog post so I'll answer here.  I used Build 656 of the OLPC software, which includes kernel version 2.6.22-20071231.2.olpc.83e0631da83a269.  In case you're interested, you can find the source code for that kernel at http://dev.laptop.org/~dilinger/stable/kernel-2.6.22-20071231.2.olpc.83e0631da83a269.src.rpm.

The only difference between that kernel and the one that shipped with the Give One Get One XOs (the kernel in Build 650) is a fix for the touchpad so you'll probably get the same results whether you're using Build 650, 653, or 656.  I don't expect that updated user applications would affect the performance of the SD card reader.

Denver

[erj]

Here is my two cents worth on the flash device speeds based on the devices I have on hand, and with the addition of boot timings.  My XO is not overclocked.  Only non-destructive reads are done in these tests.  Writes are usually slower.

The boot timing is based on the timing information displayed on the left margin during the boot process.  The value used is the last time displayed before the "Hello, (Children of the) World" message.  This method is  very "real world".  Of course, not all the time is spent doing I/O, but the variation should be mostly be due to the different devices/ports.  There appears to be a little variation in multiple trials, so I report only to the rounded tenth of a second.  The hdparm command was run with the -t option (e.g. hdparm -t /dev/sda1).  The dd command used was "dd if=/dev/{device} of=/dev/null count=1000000" to insure that buffering is not a factor.  All are averaged over 3 trials except the NAND dd tests, which used several different large files.

Read                           Boot   hdparm   dd(512MB)
From   Device                  sec    MB/sec   MB/sec   Comment
----   ------                  ----   ------   ------   -------
NAND   Internal flash          30.6   dev?     ~4.2   dd on 6 to 10 MB files (no count)
USB    PNY Optima Pro 8GB      27.8   16.2     17.8   Top right port
SD     SanDisk Extreme III 8GB 21.9    5.0      5.9   XO SD slot
USB    SanDisk Extreme III 8GB 23.8   16.7     12.3   Installed in SanDisk MicroMate reader
USB    SanDisk class 2 4GB     23.4    9.6      8.5   Installed in SanDisk MicroMate reader
SD     SanDisk class 2 4GB     22.0    4.0      4.2   XO SD slot
USB    PNY Attache 1GB         N/A    11.6     10.9   Top right port (no O/S installed)
USB    External 120GB disk     24.5   21.0     15.6   left port

What can be deduced from these results?

First, most late model external flash devices should improve performance of booting and general operation of any distribution installed on it.

Second, sustained I/O throughput is not a good predictor of faster boot times.

Third, although the SD card slot is maxed out at 5 MB/s as reported before, performance of devices in this slot can be better than expected from the sustained I/O numbers.  This is also subjectively verified by my first installing openSuSE 10.2 on the PNY Optima Pro 8GB flash drive and then about a month later cloning this install to the SanDisk Extreme III 8GB SD card.  The boot times are faster for the SD and the O/S runs noticeably smoother/faster.  It may be that average boot times will be a better predictor of O/S performance than sustained throughput for a specific device plugged into a different interface.