MFCard access speed difference between RPi and Beaglebone

[Warner Losh]

On Feb 21, 2013, at 10:02 AM, Tim Kientzle wrote:

> On Feb 21, 2013, at 6:45 AM, Werner Thie wrote:
> 
>> Hi
>> 
>> trying to nail down the crash and core dump in Python when importing ctypes I did a parallel build of python on both systems with the same SVN rev 246947 of kernel.
>> 
>> The RPi seems to beat the Bbone quite clearly with a ration of 6 : 10
>> 
>> It is very apparent (when unpacking for instance) that the Bb shows a much more pronounced 'hickup' behavior, unpacking 40-60 files and then most probably doing disk IO.
>> 
>> Swapping the MFCards does not change the situation at all.
>> 
>> Can anybody out there suppprt this observation and/or explain it?
> 
> I've seen the same behavior with things like SVN checkouts;
> the BBone runs quite smoothly for a while and then stalls.
> 
> I've been assuming that it was the MMCSD driver; skimming
> through the code, it looks like there's a lot of room for
> improvement.

I write mmcsd(4) for crappy hardware (the AT91RM9200's mmc controller was a new low in low performance), so I never got a chance to do much optimization.  I'm sure there's much low-hanging fruit there...

Warner