gumstix

Hi AJ ONeal,

Very interesting benchmark about microSDHC speed. I have tested in my Overo with a Kingston 8 GB class 4, booting from NAND (Kernel 2.6.33) :

root@...:~# dd if=/dev/urandom of=/media/mmcblk0p1/rand.dd bs=1M count=100
100+0 records in
100+0 records out
104857600 bytes (105 MB) copied, 189.881 s, 552 kB/s

552 kB/s writing over it  seems to be a little bit slow, right? (I'm not sure if I have run 'dd' correctly)

Víctor Andrés

  ----- Original Message ----- 
  From: AJ ONeal 
  To: General mailing list for gumstix users. 
  Sent: Friday, September 03, 2010 2:25 AM
  Subject: [Gumstix-users] microSDHC write speed comparison


  I'm testing out various brands and classes of microSDHC to see which actually perform as they should and whether the microSDHC or the gumstix data bus is my write speed bottleneck.


  http://fastr.github.com/articles/testing-file-write-speeds.html


  So far I would recommend Kingston. It's more expensive, but it performs well. I should have a class 10 of theirs soon to test against.

  AJ ONeal



------------------------------------------------------------------------------


  ------------------------------------------------------------------------------
  This SF.net Dev2Dev email is sponsored by:

  Show off your parallel programming skills.
  Enter the Intel(R) Threading Challenge 2010.
  http://p.sf.net/sfu/intel-thread-sfd


------------------------------------------------------------------------------


  _______________________________________________
  gumstix-users mailing list
  gumstix-users@...
  https://lists.sourceforge.net/lists/listinfo/gumstix-users

> root@...:~# dd if=/dev/urandom of=/media/mmcblk0p1/rand.dd bs=1M
> count=100
>

If your input is urandom the bottleneck could be the random number generator
and not the disk.


>
>
> ----- Original Message -----
> *From:* AJ ONeal <coolaj86@...>
> *To:* General mailing list for gumstix users.<gumstix-users@...>
> *Sent:* Friday, September 03, 2010 2:25 AM
> *Subject:* [Gumstix-users] microSDHC write speed comparison
>
> I'm testing out various brands and classes of microSDHC to see which
> actually perform as they should and whether the microSDHC or the gumstix
> data bus is my write speed bottleneck.
>
> http://fastr.github.com/articles/testing-file-write-speeds.html
>
> So far I would recommend Kingston. It's more expensive, but it performs
> well. I should have a class 10 of theirs soon to test against.
>
> AJ ONeal
>
> ------------------------------
>
>
> ------------------------------------------------------------------------------
> This SF.net Dev2Dev email is sponsored by:
>
> Show off your parallel programming skills.
> Enter the Intel(R) Threading Challenge 2010.
> http://p.sf.net/sfu/intel-thread-sfd
>
> ------------------------------
>
> _______________________________________________
> gumstix-users mailing list
> gumstix-users@...
> https://lists.sourceforge.net/lists/listinfo/gumstix-users
>
>
>
> ------------------------------------------------------------------------------
> This SF.net Dev2Dev email is sponsored by:
>
> Show off your parallel programming skills.
> Enter the Intel(R) Threading Challenge 2010.
> http://p.sf.net/sfu/intel-thread-sfd
> _______________________________________________
> gumstix-users mailing list
> gumstix-users@...
> https://lists.sourceforge.net/lists/listinfo/gumstix-users
>
>

Hi Cole,

Agreed.  I have some numbers for various random number generators on a
16-core Itanium (RHEL 4):

Method	Speed (MB/s)	Speed (MB/s)
		(saving data)	(pure generation)
Urandom	4.19	4.61
System	39.67	250.94
Cran32	37.64	186.29
Lcg	29.16	72.98
Rtprng	40.17	276.17
Isaac	42.07	375.38
Sfmt	43.32	563.38
TR1-MT	29.66	83.17

As you can see, urandom is by far the slowest, followed by random
(random blocks waiting for entropy if necessary, making it quite hard
to get lots of data).  Of course, for write testing, one doesn't even
need random data...


Robert Vogt IV
CEO
IOSiX, LLC
2375 Parkwood Ave
Ypsilanti, MI  48198
robert@...
P: 734-730-9690
F: 734-482-2337



On Fri, Sep 3, 2010 at 4:49 AM, Cole Christensen <chri2586@...> wrote:
>
>> root@...:~# dd if=/dev/urandom of=/media/mmcblk0p1/rand.dd bs=1M
>> count=100
>
> If your input is urandom the bottleneck could be the random number generator
> and not the disk.
>
>>
>>
>>
>> ----- Original Message -----
>> From: AJ ONeal
>> To: General mailing list for gumstix users.
>> Sent: Friday, September 03, 2010 2:25 AM
>> Subject: [Gumstix-users] microSDHC write speed comparison
>> I'm testing out various brands and classes of microSDHC to see which
>> actually perform as they should and whether the microSDHC or the gumstix
>> data bus is my write speed bottleneck.
>> http://fastr.github.com/articles/testing-file-write-speeds.html
>> So far I would recommend Kingston. It's more expensive, but it performs
>> well. I should have a class 10 of theirs soon to test against.
>> AJ ONeal
>>
>> ________________________________
>>
>>
>> ------------------------------------------------------------------------------
>> This SF.net Dev2Dev email is sponsored by:
>>
>> Show off your parallel programming skills.
>> Enter the Intel(R) Threading Challenge 2010.
>> http://p.sf.net/sfu/intel-thread-sfd
>>
>> ________________________________
>>
>> _______________________________________________
>> gumstix-users mailing list
>> gumstix-users@...
>> https://lists.sourceforge.net/lists/listinfo/gumstix-users
>>
>>
>> ------------------------------------------------------------------------------
>> This SF.net Dev2Dev email is sponsored by:
>>
>> Show off your parallel programming skills.
>> Enter the Intel(R) Threading Challenge 2010.
>> http://p.sf.net/sfu/intel-thread-sfd
>> _______________________________________________
>> gumstix-users mailing list
>> gumstix-users@...
>> https://lists.sourceforge.net/lists/listinfo/gumstix-users
>>
>
>
> ------------------------------------------------------------------------------
> This SF.net Dev2Dev email is sponsored by:
>
> Show off your parallel programming skills.
> Enter the Intel(R) Threading Challenge 2010.
> http://p.sf.net/sfu/intel-thread-sfd
> _______________________________________________
> gumstix-users mailing list
> gumstix-users@...
> https://lists.sourceforge.net/lists/listinfo/gumstix-users
>
>

AJ,

I believe your test method is flawed.

My results for a (generic) Class 4 MicroSD on Overo, booted from the SD
card:

root@...:/# dd if=/dev/zero of=/zero.dat bs=1M count=1000
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB) copied, 190.283 s, 5.5 MB/s

while your cat | dd is still reading from the disk, hence the slowdown.

Yours is a test of copy speed, not pure write speed.
I am not yet sure if my method is entirely valid either.
Would transferring speed be artificially high when all the bits are zero?

On Fri, Sep 3, 2010 at 11:13 AM, Robert Vogt IV <robert@...> wrote:

> Hi Cole,
>
> Agreed.  I have some numbers for various random number generators on a
> 16-core Itanium (RHEL 4):
>
> Method  Speed (MB/s)    Speed (MB/s)
>                (saving data)   (pure generation)
> Urandom 4.19    4.61
> System  39.67   250.94
> Cran32  37.64   186.29
> Lcg     29.16   72.98
> Rtprng  40.17   276.17
> Isaac   42.07   375.38
> Sfmt    43.32   563.38
> TR1-MT  29.66   83.17
>
> As you can see, urandom is by far the slowest, followed by random
> (random blocks waiting for entropy if necessary, making it quite hard
> to get lots of data).  Of course, for write testing, one doesn't even
> need random data...
>
>
> Robert Vogt IV
> CEO
> IOSiX, LLC
> 2375 Parkwood Ave
> Ypsilanti, MI  48198
> robert@...
> P: 734-730-9690
> F: 734-482-2337
>
>
>
> On Fri, Sep 3, 2010 at 4:49 AM, Cole Christensen <chri2586@...> wrote:
> >
> >> root@...:~# dd if=/dev/urandom of=/media/mmcblk0p1/rand.dd bs=1M
> >> count=100
> >
> > If your input is urandom the bottleneck could be the random number
> generator
> > and not the disk.
> >
> >>
> >>
> >>
> >> ----- Original Message -----
> >> From: AJ ONeal
> >> To: General mailing list for gumstix users.
> >> Sent: Friday, September 03, 2010 2:25 AM
> >> Subject: [Gumstix-users] microSDHC write speed comparison
> >> I'm testing out various brands and classes of microSDHC to see which
> >> actually perform as they should and whether the microSDHC or the gumstix
> >> data bus is my write speed bottleneck.
> >> http://fastr.github.com/articles/testing-file-write-speeds.html
> >> So far I would recommend Kingston. It's more expensive, but it performs
> >> well. I should have a class 10 of theirs soon to test against.
> >> AJ ONeal
> >>
> >> ________________________________
> >>
> >>
> >>
> ------------------------------------------------------------------------------
> >> This SF.net Dev2Dev email is sponsored by:
> >>
> >> Show off your parallel programming skills.
> >> Enter the Intel(R) Threading Challenge 2010.
> >> http://p.sf.net/sfu/intel-thread-sfd
> >>
> >> ________________________________
> >>
> >> _______________________________________________
> >> gumstix-users mailing list
> >> gumstix-users@...
> >> https://lists.sourceforge.net/lists/listinfo/gumstix-users
> >>
> >>
> >>
> ------------------------------------------------------------------------------
> >> This SF.net Dev2Dev email is sponsored by:
> >>
> >> Show off your parallel programming skills.
> >> Enter the Intel(R) Threading Challenge 2010.
> >> http://p.sf.net/sfu/intel-thread-sfd
> >> _______________________________________________
> >> gumstix-users mailing list
> >> gumstix-users@...
> >> https://lists.sourceforge.net/lists/listinfo/gumstix-users
> >>
> >
> >
> >
> ------------------------------------------------------------------------------
> > This SF.net Dev2Dev email is sponsored by:
> >
> > Show off your parallel programming skills.
> > Enter the Intel(R) Threading Challenge 2010.
> > http://p.sf.net/sfu/intel-thread-sfd
> > _______________________________________________
> > gumstix-users mailing list
> > gumstix-users@...
> > https://lists.sourceforge.net/lists/listinfo/gumstix-users
> >
> >
>
>
> ------------------------------------------------------------------------------
> This SF.net Dev2Dev email is sponsored by:
>
> Show off your parallel programming skills.
> Enter the Intel(R) Threading Challenge 2010.
> http://p.sf.net/sfu/intel-thread-sfd
> _______________________________________________
> gumstix-users mailing list
> gumstix-users@...
> https://lists.sourceforge.net/lists/listinfo/gumstix-users
>

Ah, but it isn't.

Notice that I'm not using the disk. I'm using memory.

*/dev/shm is memory.*


I'm not sure about zero bits.
Normally I would say no... but you never know what optimization techniques
might be employed under the hood.
Engineers are getting smarter all the time - think of how LCDs consumer less
power when the screen has a colorful image than when it is all black pixels.


*What brand of SD card are you using?*

I think the KingMax brand is akin to the chinese MP4 players you see on
ebay. It's not really class 10, it just says that. Just look at the price on
amazon. It wouldn't be so cheap if it were a real class 10.

I have a Kingston class 10 on order. I'll update when I get that.



Another thing to factor in is that *`dd` exists when the write finishes to
the buffer, not disk.
*This modification would fix that:

    #NOTE: This fills the cache buffers with data
    cat /dev/shm/rand.dd /dev/shm/rand.dd | dd of=${MMC}/rand.dd

    # The overhead of the un`sync`d data above will make this more accurate
    cat /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd \
      /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd \
      /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd /dev/shm/rand.dd \
      ....


AJ ONeal


On Fri, Sep 3, 2010 at 11:35 AM, Cole Christensen <chri2586@...> wrote:

> AJ,
>
> I believe your test method is flawed.
>
> My results for a (generic) Class 4 MicroSD on Overo, booted from the SD
> card:
>
> root@...:/# dd if=/dev/zero of=/zero.dat bs=1M count=1000
> 1000+0 records in
> 1000+0 records out
> 1048576000 bytes (1.0 GB) copied, 190.283 s, 5.5 MB/s
>
> while your cat | dd is still reading from the disk, hence the slowdown.
>
> Yours is a test of copy speed, not pure write speed.
> I am not yet sure if my method is entirely valid either.
> Would transferring speed be artificially high when all the bits are zero?
>
>
> On Fri, Sep 3, 2010 at 11:13 AM, Robert Vogt IV <robert@...> wrote:
>
>> Hi Cole,
>>
>> Agreed.  I have some numbers for various random number generators on a
>> 16-core Itanium (RHEL 4):
>>
>> Method  Speed (MB/s)    Speed (MB/s)
>>                (saving data)   (pure generation)
>> Urandom 4.19    4.61
>> System  39.67   250.94
>> Cran32  37.64   186.29
>> Lcg     29.16   72.98
>> Rtprng  40.17   276.17
>> Isaac   42.07   375.38
>> Sfmt    43.32   563.38
>> TR1-MT  29.66   83.17
>>
>> As you can see, urandom is by far the slowest, followed by random
>> (random blocks waiting for entropy if necessary, making it quite hard
>> to get lots of data).  Of course, for write testing, one doesn't even
>> need random data...
>>
>>
>> Robert Vogt IV
>> CEO
>> IOSiX, LLC
>> 2375 Parkwood Ave
>> Ypsilanti, MI  48198
>> robert@...
>> P: 734-730-9690
>> F: 734-482-2337
>>
>>
>>
>> On Fri, Sep 3, 2010 at 4:49 AM, Cole Christensen <chri2586@...>
>> wrote:
>> >
>> >> root@...:~# dd if=/dev/urandom of=/media/mmcblk0p1/rand.dd bs=1M
>> >> count=100
>> >
>> > If your input is urandom the bottleneck could be the random number
>> generator
>> > and not the disk.
>> >
>> >>
>> >>
>> >>
>> >> ----- Original Message -----
>> >> From: AJ ONeal
>> >> To: General mailing list for gumstix users.
>> >> Sent: Friday, September 03, 2010 2:25 AM
>> >> Subject: [Gumstix-users] microSDHC write speed comparison
>> >> I'm testing out various brands and classes of microSDHC to see which
>> >> actually perform as they should and whether the microSDHC or the
>> gumstix
>> >> data bus is my write speed bottleneck.
>> >> http://fastr.github.com/articles/testing-file-write-speeds.html
>> >> So far I would recommend Kingston. It's more expensive, but it performs
>> >> well. I should have a class 10 of theirs soon to test against.
>> >> AJ ONeal
>> >>
>> >> ________________________________
>> >>
>> >>
>> >>
>> ------------------------------------------------------------------------------
>> >> This SF.net Dev2Dev email is sponsored by:
>> >>
>> >> Show off your parallel programming skills.
>> >> Enter the Intel(R) Threading Challenge 2010.
>> >> http://p.sf.net/sfu/intel-thread-sfd
>> >>
>> >> ________________________________
>> >>
>> >> _______________________________________________
>> >> gumstix-users mailing list
>> >> gumstix-users@...
>> >> https://lists.sourceforge.net/lists/listinfo/gumstix-users
>> >>
>> >>
>> >>
>> ------------------------------------------------------------------------------
>> >> This SF.net Dev2Dev email is sponsored by:
>> >>
>> >> Show off your parallel programming skills.
>> >> Enter the Intel(R) Threading Challenge 2010.
>> >> http://p.sf.net/sfu/intel-thread-sfd
>> >> _______________________________________________
>> >> gumstix-users mailing list
>> >> gumstix-users@...
>> >> https://lists.sourceforge.net/lists/listinfo/gumstix-users
>> >>
>> >
>> >
>> >
>> ------------------------------------------------------------------------------
>> > This SF.net Dev2Dev email is sponsored by:
>> >
>> > Show off your parallel programming skills.
>> > Enter the Intel(R) Threading Challenge 2010.
>> > http://p.sf.net/sfu/intel-thread-sfd
>> > _______________________________________________
>> > gumstix-users mailing list
>> > gumstix-users@...
>> > https://lists.sourceforge.net/lists/listinfo/gumstix-users
>> >
>> >
>>
>>
>> ------------------------------------------------------------------------------
>> This SF.net Dev2Dev email is sponsored by:
>>
>> Show off your parallel programming skills.
>> Enter the Intel(R) Threading Challenge 2010.
>> http://p.sf.net/sfu/intel-thread-sfd
>> _______________________________________________
>> gumstix-users mailing list
>> gumstix-users@...
>> https://lists.sourceforge.net/lists/listinfo/gumstix-users
>>
>
>
>
> ------------------------------------------------------------------------------
> This SF.net Dev2Dev email is sponsored by:
>
> Show off your parallel programming skills.
> Enter the Intel(R) Threading Challenge 2010.
> http://p.sf.net/sfu/intel-thread-sfd
> _______________________________________________
> gumstix-users mailing list
> gumstix-users@...
> https://lists.sourceforge.net/lists/listinfo/gumstix-users
>
>

That said, there may be some delay as data is transferred from memory.

When you `dd if=/dev/zero` I think the CPU just spits zeros at the disk.

I would imagine that the overhead of memory could be part of the 5.5 /
5.3 discrepancy.

You also may have a particularly good card.


AJ ONeal

>From a microSD standpoint, you have to send the zeros over SPI and
they have to get written to the flash memory, so there should be no
optimization for that...
Hi AJ,

And I wouldn't call 5.3-5.5 MB/s a discrepancy.  inodes have to be
written and adjusted and sectors allocated as the write progresses.
Certainly this depends on how freshly formatted the card is, the exact
size, etc.


Robert Vogt IV
CEO
IOSiX, LLC
2375 Parkwood Ave
Ypsilanti, MI  48198
robert@...
P: 734-730-9690
F: 734-482-2337



On Fri, Sep 3, 2010 at 2:05 PM, AJ ONeal <coolaj86@...> wrote:
> That said, there may be some delay as data is transferred from memory.
> When you `dd if=/dev/zero` I think the CPU just spits zeros at the disk.
> I would imagine that the overhead of memory could be part of the 5.5 /
> 5.3 discrepancy.
> You also may have a particularly good card.
>
>
> AJ ONeal
>
> ------------------------------------------------------------------------------
> This SF.net Dev2Dev email is sponsored by:
>
> Show off your parallel programming skills.
> Enter the Intel(R) Threading Challenge 2010.
> http://p.sf.net/sfu/intel-thread-sfd
> _______________________________________________
> gumstix-users mailing list
> gumstix-users@...
> https://lists.sourceforge.net/lists/listinfo/gumstix-users
>
>

>From a Kingston 2GB class 4 on Overo, booted from the microSD card:

root@...:~# dd if=/dev/zero of=/zero.dat bs=1M count=1000
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB) copied, 236.848 s, 4.4 MB/s

There's about 1.1 MB/s different with the result of your experience.


But there's something that I don't understand, if I boot from NAND and run the same I get only 1.2  MB/s.

root@...:~# dd if=/dev/zero of=/media/mmcblk0p1/zero.dat bs=1M count=1000
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB) copied, 859.527 s, 1.2 MB/s


Regards

  ----- Original Message ----- 
  From: Cole Christensen 
  To: General mailing list for gumstix users. 
  Sent: Friday, September 03, 2010 7:35 PM
  Subject: Re: [Gumstix-users] microSDHC write speed comparison


  AJ,

  I believe your test method is flawed.

  My results for a (generic) Class 4 MicroSD on Overo, booted from the SD card:

  root@...:/# dd if=/dev/zero of=/zero.dat bs=1M count=1000
  1000+0 records in
  1000+0 records out
  1048576000 bytes (1.0 GB) copied, 190.283 s, 5.5 MB/s

  while your cat | dd is still reading from the disk, hence the slowdown.

  Yours is a test of copy speed, not pure write speed.  
  I am not yet sure if my method is entirely valid either. 
  Would transferring speed be artificially high when all the bits are zero?


  On Fri, Sep 3, 2010 at 11:13 AM, Robert Vogt IV <robert@...> wrote:

    Hi Cole,

    Agreed.  I have some numbers for various random number generators on a
    16-core Itanium (RHEL 4):

    Method  Speed (MB/s)    Speed (MB/s)
                   (saving data)   (pure generation)
    Urandom 4.19    4.61
    System  39.67   250.94
    Cran32  37.64   186.29
    Lcg     29.16   72.98
    Rtprng  40.17   276.17
    Isaac   42.07   375.38
    Sfmt    43.32   563.38
    TR1-MT  29.66   83.17

    As you can see, urandom is by far the slowest, followed by random
    (random blocks waiting for entropy if necessary, making it quite hard
    to get lots of data).  Of course, for write testing, one doesn't even
    need random data...


    Robert Vogt IV
    CEO
    IOSiX, LLC
    2375 Parkwood Ave
    Ypsilanti, MI  48198
    robert@...
    P: 734-730-9690
    F: 734-482-2337




    On Fri, Sep 3, 2010 at 4:49 AM, Cole Christensen <chri2586@...> wrote:
    >
    >> root@...:~# dd if=/dev/urandom of=/media/mmcblk0p1/rand.dd bs=1M
    >> count=100
    >
    > If your input is urandom the bottleneck could be the random number generator
    > and not the disk.
    >
    >>
    >>
    >>
    >> ----- Original Message -----
    >> From: AJ ONeal
    >> To: General mailing list for gumstix users.
    >> Sent: Friday, September 03, 2010 2:25 AM
    >> Subject: [Gumstix-users] microSDHC write speed comparison
    >> I'm testing out various brands and classes of microSDHC to see which
    >> actually perform as they should and whether the microSDHC or the gumstix
    >> data bus is my write speed bottleneck.
    >> http://fastr.github.com/articles/testing-file-write-speeds.html
    >> So far I would recommend Kingston. It's more expensive, but it performs
    >> well. I should have a class 10 of theirs soon to test against.
    >> AJ ONeal
    >>
    >> ________________________________
    >>
    >>
    >> ------------------------------------------------------------------------------
    >> This SF.net Dev2Dev email is sponsored by:
    >>
    >> Show off your parallel programming skills.
    >> Enter the Intel(R) Threading Challenge 2010.
    >> http://p.sf.net/sfu/intel-thread-sfd
    >>
    >> ________________________________
    >>
    >> _______________________________________________
    >> gumstix-users mailing list
    >> gumstix-users@...
    >> https://lists.sourceforge.net/lists/listinfo/gumstix-users
    >>
    >>
    >> ------------------------------------------------------------------------------
    >> This SF.net Dev2Dev email is sponsored by:
    >>
    >> Show off your parallel programming skills.
    >> Enter the Intel(R) Threading Challenge 2010.
    >> http://p.sf.net/sfu/intel-thread-sfd
    >> _______________________________________________
    >> gumstix-users mailing list
    >> gumstix-users@...
    >> https://lists.sourceforge.net/lists/listinfo/gumstix-users
    >>
    >
    >
    > ------------------------------------------------------------------------------
    > This SF.net Dev2Dev email is sponsored by:
    >
    > Show off your parallel programming skills.
    > Enter the Intel(R) Threading Challenge 2010.
    > http://p.sf.net/sfu/intel-thread-sfd
    > _______________________________________________
    > gumstix-users mailing list
    > gumstix-users@...
    > https://lists.sourceforge.net/lists/listinfo/gumstix-users
    >
    >

    ------------------------------------------------------------------------------
    This SF.net Dev2Dev email is sponsored by:

    Show off your parallel programming skills.
    Enter the Intel(R) Threading Challenge 2010.
    http://p.sf.net/sfu/intel-thread-sfd
    _______________________________________________
    gumstix-users mailing list
    gumstix-users@...
    https://lists.sourceforge.net/lists/listinfo/gumstix-users





------------------------------------------------------------------------------


  ------------------------------------------------------------------------------
  This SF.net Dev2Dev email is sponsored by:

  Show off your parallel programming skills.
  Enter the Intel(R) Threading Challenge 2010.
  http://p.sf.net/sfu/intel-thread-sfd


------------------------------------------------------------------------------


  _______________________________________________
  gumstix-users mailing list
  gumstix-users@...
  https://lists.sourceforge.net/lists/listinfo/gumstix-users

What if you try copying from memory rather than dd if=/dev/zero?

It may be that when you boot from NAND it's automatically mounted in 'sync'
mode rather than 'async'.

AJ ONeal

On Mon, Sep 6, 2010 at 4:10 AM, Victor Andres <victor@...> wrote:

>  From a Kingston 2GB class 4 on Overo, booted from the microSD card:
>
> root@...:~# dd if=/dev/zero of=/zero.dat bs=1M count=1000
> 1000+0 records in
> 1000+0 records out
> 1048576000 bytes (1.0 GB) copied, 236.848 s, 4.4 MB/s
>
> There's about 1.1 MB/s different with the result of your experience.
>
>
> But there's something that I don't understand, if I boot from NAND and run
> the same I get only 1.2  MB/s.
>
> root@...:~# dd if=/dev/zero of=/media/mmcblk0p1/zero.dat bs=1M count=1000
>
> 1000+0 records in
> 1000+0 records out
> 1048576000 bytes (1.0 GB) copied, 859.527 s, 1.2 MB/s
>
> Regards
>
>
> ----- Original Message -----
> *From:* Cole Christensen <chri2586@...>
> *To:* General mailing list for gumstix users.<gumstix-users@...>
> *Sent:* Friday, September 03, 2010 7:35 PM
> *Subject:* Re: [Gumstix-users] microSDHC write speed comparison
>
> AJ,
>
> I believe your test method is flawed.
>
> My results for a (generic) Class 4 MicroSD on Overo, booted from the SD
> card:
>
> root@...:/# dd if=/dev/zero of=/zero.dat bs=1M count=1000
> 1000+0 records in
> 1000+0 records out
> 1048576000 bytes (1.0 GB) copied, 190.283 s, 5.5 MB/s
>
> while your cat | dd is still reading from the disk, hence the slowdown.
>
> Yours is a test of copy speed, not pure write speed.
> I am not yet sure if my method is entirely valid either.
> Would transferring speed be artificially high when all the bits are zero?
>
> On Fri, Sep 3, 2010 at 11:13 AM, Robert Vogt IV <robert@...> wrote:
>
>> Hi Cole,
>>
>> Agreed.  I have some numbers for various random number generators on a
>> 16-core Itanium (RHEL 4):
>>
>> Method  Speed (MB/s)    Speed (MB/s)
>>                (saving data)   (pure generation)
>> Urandom 4.19    4.61
>> System  39.67   250.94
>> Cran32  37.64   186.29
>> Lcg     29.16   72.98
>> Rtprng  40.17   276.17
>> Isaac   42.07   375.38
>> Sfmt    43.32   563.38
>> TR1-MT  29.66   83.17
>>
>> As you can see, urandom is by far the slowest, followed by random
>> (random blocks waiting for entropy if necessary, making it quite hard
>> to get lots of data).  Of course, for write testing, one doesn't even
>> need random data...
>>
>>
>> Robert Vogt IV
>> CEO
>> IOSiX, LLC
>> 2375 Parkwood Ave
>> Ypsilanti, MI  48198
>> robert@...
>> P: 734-730-9690
>> F: 734-482-2337
>>
>>
>>
>> On Fri, Sep 3, 2010 at 4:49 AM, Cole Christensen <chri2586@...>
>> wrote:
>> >
>> >> root@...:~# dd if=/dev/urandom of=/media/mmcblk0p1/rand.dd bs=1M
>> >> count=100
>> >
>> > If your input is urandom the bottleneck could be the random number
>> generator
>> > and not the disk.
>> >
>> >>
>> >>
>> >>
>> >> ----- Original Message -----
>> >> From: AJ ONeal
>> >> To: General mailing list for gumstix users.
>> >> Sent: Friday, September 03, 2010 2:25 AM
>> >> Subject: [Gumstix-users] microSDHC write speed comparison
>> >> I'm testing out various brands and classes of microSDHC to see which
>> >> actually perform as they should and whether the microSDHC or the
>> gumstix
>> >> data bus is my write speed bottleneck.
>> >> http://fastr.github.com/articles/testing-file-write-speeds.html
>> >> So far I would recommend Kingston. It's more expensive, but it performs
>> >> well. I should have a class 10 of theirs soon to test against.
>> >> AJ ONeal
>> >>
>> >> ________________________________
>> >>
>> >>
>> >>
>> ------------------------------------------------------------------------------
>> >> This SF.net Dev2Dev email is sponsored by:
>> >>
>> >> Show off your parallel programming skills.
>> >> Enter the Intel(R) Threading Challenge 2010.
>> >> http://p.sf.net/sfu/intel-thread-sfd
>> >>
>> >> ________________________________
>> >>
>> >> _______________________________________________
>> >> gumstix-users mailing list
>> >> gumstix-users@...
>> >> https://lists.sourceforge.net/lists/listinfo/gumstix-users
>> >>
>> >>
>> >>
>> ------------------------------------------------------------------------------
>> >> This SF.net Dev2Dev email is sponsored by:
>> >>
>> >> Show off your parallel programming skills.
>> >> Enter the Intel(R) Threading Challenge 2010.
>> >> http://p.sf.net/sfu/intel-thread-sfd
>> >> _______________________________________________
>> >> gumstix-users mailing list
>> >> gumstix-users@...
>> >> https://lists.sourceforge.net/lists/listinfo/gumstix-users
>> >>
>> >
>> >
>> >
>> ------------------------------------------------------------------------------
>> > This SF.net Dev2Dev email is sponsored by:
>> >
>> > Show off your parallel programming skills.
>> > Enter the Intel(R) Threading Challenge 2010.
>> > http://p.sf.net/sfu/intel-thread-sfd
>> > _______________________________________________
>> > gumstix-users mailing list
>> > gumstix-users@...
>> > https://lists.sourceforge.net/lists/listinfo/gumstix-users
>> >
>> >
>>
>>
>> ------------------------------------------------------------------------------
>> This SF.net Dev2Dev email is sponsored by:
>>
>> Show off your parallel programming skills.
>> Enter the Intel(R) Threading Challenge 2010.
>> http://p.sf.net/sfu/intel-thread-sfd
>> _______________________________________________
>> gumstix-users mailing list
>> gumstix-users@...
>> https://lists.sourceforge.net/lists/listinfo/gumstix-users
>>
>
>  ------------------------------
>
>
> ------------------------------------------------------------------------------
> This SF.net Dev2Dev email is sponsored by:
>
> Show off your parallel programming skills.
> Enter the Intel(R) Threading Challenge 2010.
> http://p.sf.net/sfu/intel-thread-sfd
>
> ------------------------------
>
> _______________________________________________
> gumstix-users mailing list
> gumstix-users@...
> https://lists.sourceforge.net/lists/listinfo/gumstix-users
>
>
>
> ------------------------------------------------------------------------------
> This SF.net Dev2Dev email is sponsored by:
>
> Show off your parallel programming skills.
> Enter the Intel(R) Threading Challenge 2010.
> http://p.sf.net/sfu/intel-thread-sfd
> _______________________________________________
> gumstix-users mailing list
> gumstix-users@...
> https://lists.sourceforge.net/lists/listinfo/gumstix-users
>
>

I got the Kingston 16GB class 10 and it's hanging right around 4.9MB/s write
speed as well.

Having tested two class 10s, I believe the fault may lie with the Gumstix
itself.

Nevertheless, 4.5MB/s is good enough for my application - if it is
consistent even with other system loads.

For reference:
the network transfer rate is about 10MB/s (100mb/s), but less using ssh
(encryption slows the transfer).

AJ ONeal

On Thu, Sep 2, 2010 at 6:25 PM, AJ ONeal <coolaj86@...> wrote:

> I'm testing out various brands and classes of microSDHC to see which
> actually perform as they should and whether the microSDHC or the gumstix
> data bus is my write speed bottleneck.
>
> http://fastr.github.com/articles/testing-file-write-speeds.html
>
> So far I would recommend Kingston. It's more expensive, but it performs
> well. I should have a class 10 of theirs soon to test against.
>
> AJ ONeal
>

Hi AJ,

Just got around to running some tests with various cards I have here.
Booted into NAND via 'run nandboot' in uboot.


This is with a fresh ext3 filesystem on a '3C Pro' 16GB class 6 microSDHC card:

root@...:~# dd if=/dev/zero of=/media/data/zero.dat bs=10M count=100
100+0 records in
100+0 records out
1048576000 bytes (1.0 GB) copied, 290.395 s, 3.6 MB/s
root@...:~# dd if=/dev/zero of=/media/data/zero.dat bs=1M count=1000
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB) copied, 282.553 s, 3.7 MB/s


With a fresh ext3 filesystem on a 'PNY' 8GB class 4 microSDHC card:

root@...# dd if=/dev/zero of=/media/mmcblk0p2/zero.dat bs=10M count=100
100+0 records in
100+0 records out
1048576000 bytes (1.0 GB) copied, 310.917 s, 3.4 MB/s
root@...# dd if=/dev/zero of=/media/mmcblk0p2/zero.dat bs=1M count=1000
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB) copied, 900.03 s, 1.2 MB/s
root@...# dd if=/dev/zero of=/media/mmcblk0p2/zero.dat bs=1M count=1000
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB) copied, 903.892 s, 1.2 MB/s
root@...# dd if=/dev/zero of=/media/mmcblk0p2/zero.dat bs=10M count=100
100+0 records in
100+0 records out
1048576000 bytes (1.0 GB) copied, 294.93 s, 3.6 MB/s
[kind of a weird discrepancy with the blocksize...]


With a fresh ext3 filesystem on a generic 2GB microSD card with no
class indication:

root@...# dd if=/dev/zero of=/media/mmcblk0p2/zero.dat bs=10M count=100
100+0 records in
100+0 records out
1048576000 bytes (1.0 GB) copied, 217.56 s, 4.8 MB/s
root@...# dd if=/dev/zero of=/media/mmcblk0p2/zero.dat bs=1M count=1000
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB) copied, 214.329 s, 4.9 MB/s


I was unable to get the Overo to read or write an ext4 filesystem.


Robert Vogt IV
CEO
IOSiX, LLC
2375 Parkwood Ave
Ypsilanti, MI  48198
robert@...
P: 734-730-9690
F: 734-482-2337



On Thu, Sep 2, 2010 at 8:25 PM, AJ ONeal <coolaj86@...> wrote:
> I'm testing out various brands and classes of microSDHC to see which
> actually perform as they should and whether the microSDHC or the gumstix
> data bus is my write speed bottleneck.
> http://fastr.github.com/articles/testing-file-write-speeds.html
> So far I would recommend Kingston. It's more expensive, but it performs
> well. I should have a class 10 of theirs soon to test against.
> AJ ONeal
>
> ------------------------------------------------------------------------------
> This SF.net Dev2Dev email is sponsored by:
>
> Show off your parallel programming skills.
> Enter the Intel(R) Threading Challenge 2010.
> http://p.sf.net/sfu/intel-thread-sfd
> _______________________________________________
> gumstix-users mailing list
> gumstix-users@...
> https://lists.sourceforge.net/lists/listinfo/gumstix-users
>
>