Re: [PATCH v5 2/2] mmc: core: Support packed command for eMMC4.5 device

[merez@codeaurora.org]

Hi,

Our tests showed that the write packing improved the performance of the
write sequential operations:

Long write operation:
----------------------
no-packing: 15.8 MB/s
packed commands patch (both READ and WRITE packing are enabled): 23.3 MB/s

Several parallel write operations (sum of all the write throughputs):
---------------------------
no-packing: 17.1 MB/s
packed commands patch(both READ and WRITE packing are enabled): 25 MB/s

Parallel long read and long write operations (write throughput):
-----------------------------------------------------------------
no-packing: 12.2 MB/s
packed commands patch (both READ and WRITE packing are enabled): 16.3 MB/s

Parallel short read and long write operations (write throughput):
-----------------------------------------------------------------
no-packing: 15.4 MB/s
packed commands patch (both READ and WRITE packing are enabled): 16.4 MB/s

Several Parallel short read and short write operations (sum of all the
write throughputs):
--------------------------------------------------------------------------
no-packing: 12.5 MB/s
packed commands patch (both READ and WRITE packing are enabled): 15.5 MB/s


Random read and random write:
------------------------------
I checked the random read and random write IOPs by using the IOZONE
application. There was a slight degradation in the read results due to the
packing and no improvements in the write results.

The results are:

IOZONE file size of 100M:
no-packing: random read: 4675, random write: 729
packed commands patch (both READ and WRITE packing are enabled): random
read: 4557 random write: 723

IOZONE file size of 256M:
no-packing: random read: 4632, random write: 744
packed commands patch (both READ and WRITE packing are enabled): random
read: 4498, random write: 742

Thanks,
Maya Erez
Consultant for Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum

> Hi. merez.
>
> Would you share random read speed with us ?
>
> And Write speed also..
>
> Thanks.
>
> 2012/3/1  <merez@codeaurora.org>:
>>> This patch supports packed command of eMMC4.5 device.
>>> Several reads(or writes) can be grouped in packed command
>>> and all data of the individual commands can be sent in a
>>> single transfer on the bus.
>>>
>>> Signed-off-by: Seungwon Jeon <tgih.jun@samsung.com>
>>> ---
>>>  drivers/mmc/card/block.c   |  496
>>> +++++++++++++++++++++++++++++++++++++++++--
>>>  drivers/mmc/card/queue.c   |   48 ++++-
>>>  drivers/mmc/card/queue.h   |   13 ++
>>>  drivers/mmc/core/mmc_ops.c |    1 +
>>>  include/linux/mmc/core.h   |    4 +
>>>  5 files changed, 535 insertions(+), 27 deletions(-)
>>>
>> Hi,
>>
>> We ran performance tests on the packed commands patch. We found out that
>> enabling the read packing didn't improve the performance in any of the
>> scenarios we ran (see the detailed results below).
>> Therefore, we suggest to move the read packing code to a different patch
>> and approve only the write packing code for now. The read packing adds
>> complexity to the code and we don't see a point in adding it while the
>> intention is to disable it.
>>
>> Test results:
>>
>> Long read operation:
>> ----------------------
>> no-packing: 39.5 MB/s
>> packed commands patch (both READ and WRITE packing are enabled): 39.5
>> MB/s
>> packed commands patch + enabling only READ packing: 39.5 MB/s
>>
>> Several parallel read operations (sum of all the read throughputs):
>> ---------------------------
>> no-packing: 42.6 MB/s
>> packed commands patch(both READ and WRITE packing are enabled): 38 MB/s
>> packed commands patch + enabling only READ packing: 38.2 MB/s
>>
>> Parallel long read and long write operations (read throughput):
>> -----------------------------------------------------------------
>> no-packing: 23.8 MB/s
>> packed commands patch (both READ and WRITE packing are enabled): 12.6
>> MB/s
>> packed commands patch + enabling only READ packing: 12.5 MB/s
>>
>> Parallel short read and long write operations (read throughput):
>> -----------------------------------------------------------------
>> no-packing: 22.9 MB/s
>> packed commands patch (both READ and WRITE packing are enabled): 8.4
>> MB/s
>> packed commands patch + enabling only READ packing: 8.6 MB/s
>>
>> Several Parallel short read and short write operations (sum of all the
>> read throughputs):
>> --------------------------------------------------------------------------
>> no-packing: 41.6 MB/s
>> packed commands patch (both READ and WRITE packing are enabled): 35 MB/s
>> packed commands patch + enabling only READ packing: 36 MB/s
>>
>> Thanks,
>> Maya Erez
>> Consultant for Qualcomm Innovation Center, Inc.
>> Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum
>>
>>
>>
>>
>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-mmc" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> --
> To unsubscribe from this list: send the line "unsubscribe linux-mmc" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>