ext2/3/4 group size

Yioryos Asprobounitis mavrothal at yahoo.com
Sat Apr 2 17:15:07 EDT 2011

--- On Fri, 4/1/11, Arnd Bergmann <arnd at arndb.de> wrote:

> From: Arnd Bergmann <arnd at arndb.de>
> Subject: Re: ext2/3/4 group size
> To: "Yioryos Asprobounitis" <mavrothal at yahoo.com>
> Cc: devel at lists.laptop.org
> Date: Friday, April 1, 2011, 4:14 AM
> On Thursday 31 March 2011, Yioryos
> Asprobounitis wrote:
> > --- On Thu, 3/31/11, Arnd Bergmann <arnd at arndb.de>
> wrote:
> > 
> > > From: Arnd Bergmann <arnd at arndb.de>
> > 
> > > The Kingston card probably uses a toshiba
> controller (oemid 0x544d) 
> > [snip] 
> > > The transcend cards I've seen typically use
> samsung
> > > controllers (oemid 0x534d),
>> > Here is what I get for my cards:
> > 
> > A-data          fwrev 0x0
> hwrev 0x0 manfid 0x00001d oemid 0x4144
> > Kingston        fwrev 0x0 hwrev
> 0x2 manfid 0x000041 oemid 0x3432
> > Transcend       fwrev 0x0
> hwrev 0x1 manfid 0x00001c oemid 0x5356
> > 
> > Any idea about the controllers?
> I've seen the 0x41/0x3432 before on a Kingston card, and
> this looks
> like an improper OEM ID because it's not actually letters
> in ASCII.
> Kingston was not able to confim this card to be either
> authentic
> or fake, but measurements have shown that it has the same
> flawed
> GC algorithm as the others (only one allocation unit open
> at a time,
> no random write optimizations).
> I've seen the 0x1d/0x4144 ID on cards with very different
> characteristics,
> a very good one branded "Microcenter" and a rather bad one
> branded
> "Extrememory Performance Class 6". Note that 0x4144 is "AD"
> in ASCII,
> so that is likely to actually stand for A-DATA, who would
> produce
> the controllers and sell them to the other OEMs.
> The Transcend controller is new to me, the few other
> transcend ones
> I've seen were using Samsung (0x1b/0x534D, "SM"). 0x5356 is
> "SV", which
> I have never seen before, it could plausibly stand for
> "Sunvision Co.",
> who are listed as a SD licensee but not a brand name.
> If you want to do more tests, please get the flashbench
> source code from
> "git clone git://git.linaro.org/people/arnd/flashbench.git"
> and send your
> results (one mail per card) to flashbench-results at lists.linaro.org.
> The data should include:
> * The contents of all sysfs files
> * The exact size of the card in kb
> * The output of "flashbench -a --blocksize=1024", which is
> needed to guess
>   the erase block size. The README file tells you how
> to interpret the
>   results, ask me if you're not sure.
> * When you have the erase block size, the results of
>   "flashbench --open-au --open-au-nr=${NUMBER}
> --erasesize=${ERASESIZE}"
>   for the correct erase block size and a varying
> number of erase blocks.
>   Try different numbers until you find the highest one
> that is fast
>   and the lowest one that is slow (one more than the
> first).
> * The the --open-au test with and without "--random". The
> Kingston card
>   will always be slow with --random, so no need for
> that. For cards that
>   are fast with --random, always use that, in orer to
> get more accurate
>   numbers.
> * Use the '--findfat --random' test with the correct erase
> block size to
>   determine if the card has any areas optimized for
> random access that
>   can be used to store the FAT. On the Kingston card,
> this will be the
>   second 4 MB erase block.

Thanks for the info.
I got a new Transcend 4GB class 10 card: oemid 0x4142(AB???), manfid 0x00001e(RS???), hwrev 0x1, fwrev 0x0, serial 0x00000aec 
and I checked it with flashbench when formatted as vfat (factory) 
standard ext3 (mkfs.ext3 /dev/mmcblk0p1) 
and ext3 with groups aligning to the 4MB erase block (mkfs.ext3 -b 4096 -i 4096 -g 1024 -m 2 -O ^resize_inode  /dev/mmcblk0p1)
Could not see any difference other than that in the last case the 4MB and 16KB (page size?!) boundaries where much more clear.

However, and despite the ability of the card to have 5 blocks open the card performed terribly in real life giving often I/O errors.
This was corrected when the ext3 file system was was made with 1KB, instead of the default 4KB block size, ie the `-b 1024' argument :?
However, under these conditions the card became considerably slower in real life use (though flashbench detected no difference)

All tests on an XO-1.5 running F14/os13.
attached is the flashbench data with the card formatted in 3 different ways in case someone can see something more in it.

Boy..., getting a good card becomes a lottery... 

>     Arnd

-------------- next part --------------
A non-text attachment was scrubbed...
Name: transcend_card.tar.gz
Type: application/x-gzip
Size: 3147 bytes
Desc: not available
URL: <http://lists.laptop.org/pipermail/devel/attachments/20110402/032b6d8e/attachment.bin>

More information about the Devel mailing list