Hi,
I'm doing a performance testing on my bench ARM box.
1. dd test: I have validate the read and write by mounting /dev/sda1
with ext4 filesystem,
able to get the good performance numbers where read is high
compared to write
2. robocopy test:
- mkfs.ext4 /dev/sda1
- mount /dev/sda1 /media/disk
- << configured samba >>
- Mapped the /media/disk on windows
- login on the mapped driver in windows
- did a robocopy test, where write got 84MBps and read 14MBps
read performance is too slow when compared to write in robocopy case.
Can anyone help me out, how to debug this further.
thanks!
--
Jagan.
If you are robocoping small files you will hit other limits.
Best I have seen with small files is around 30 files/second, and that
involves multiple copies going on. Remember with a small files there
are several reads and writes that need to be done to complete a create
of a small file and each of these take time. 30 files/second ~ 30ms
per file, not that bad considering that on a real spinning disk a
single read/write op is 5-10ms, and creating the file entry, copying
data and closing the file takes several operations (at least create
file entry, write small amount of data, update file entry
date/time/info). If the write in the middle is not a significant
amount of data, the 2 extra ops are what hurts.
On Fri, Nov 14, 2014 at 6:55 AM, Jagan Teki <[email protected]> wrote:
> Hi,
>
> I'm doing a performance testing on my bench ARM box.
>
> 1. dd test: I have validate the read and write by mounting /dev/sda1
> with ext4 filesystem,
> able to get the good performance numbers where read is high
> compared to write
>
> 2. robocopy test:
> - mkfs.ext4 /dev/sda1
> - mount /dev/sda1 /media/disk
> - << configured samba >>
> - Mapped the /media/disk on windows
> - login on the mapped driver in windows
> - did a robocopy test, where write got 84MBps and read 14MBps
>
> read performance is too slow when compared to write in robocopy case.
> Can anyone help me out, how to debug this further.
>
> thanks!
> --
> Jagan.
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
On 14 November 2014 18:50, Roger Heflin <[email protected]> wrote:
> If you are robocoping small files you will hit other limits.
>
> Best I have seen with small files is around 30 files/second, and that
> involves multiple copies going on. Remember with a small files there
> are several reads and writes that need to be done to complete a create
> of a small file and each of these take time. 30 files/second ~ 30ms
> per file, not that bad considering that on a real spinning disk a
> single read/write op is 5-10ms, and creating the file entry, copying
> data and closing the file takes several operations (at least create
> file entry, write small amount of data, update file entry
> date/time/info). If the write in the middle is not a significant
> amount of data, the 2 extra ops are what hurts.
>
But, I tried 4gb and 1gb files both got a similar numbers.
> On Fri, Nov 14, 2014 at 6:55 AM, Jagan Teki <[email protected]> wrote:
>> Hi,
>>
>> I'm doing a performance testing on my bench ARM box.
>>
>> 1. dd test: I have validate the read and write by mounting /dev/sda1
>> with ext4 filesystem,
>> able to get the good performance numbers where read is high
>> compared to write
>>
>> 2. robocopy test:
>> - mkfs.ext4 /dev/sda1
>> - mount /dev/sda1 /media/disk
>> - << configured samba >>
>> - Mapped the /media/disk on windows
>> - login on the mapped driver in windows
>> - did a robocopy test, where write got 84MBps and read 14MBps
>>
>> read performance is too slow when compared to write in robocopy case.
>> Can anyone help me out, how to debug this further.
thanks!
--
Jagan.
What kind of underlying disk is it?
On Fri, Nov 14, 2014 at 7:36 AM, Jagan Teki <[email protected]> wrote:
> On 14 November 2014 18:50, Roger Heflin <[email protected]> wrote:
>> If you are robocoping small files you will hit other limits.
>>
>> Best I have seen with small files is around 30 files/second, and that
>> involves multiple copies going on. Remember with a small files there
>> are several reads and writes that need to be done to complete a create
>> of a small file and each of these take time. 30 files/second ~ 30ms
>> per file, not that bad considering that on a real spinning disk a
>> single read/write op is 5-10ms, and creating the file entry, copying
>> data and closing the file takes several operations (at least create
>> file entry, write small amount of data, update file entry
>> date/time/info). If the write in the middle is not a significant
>> amount of data, the 2 extra ops are what hurts.
>>
>
> But, I tried 4gb and 1gb files both got a similar numbers.
>
>> On Fri, Nov 14, 2014 at 6:55 AM, Jagan Teki <[email protected]> wrote:
>>> Hi,
>>>
>>> I'm doing a performance testing on my bench ARM box.
>>>
>>> 1. dd test: I have validate the read and write by mounting /dev/sda1
>>> with ext4 filesystem,
>>> able to get the good performance numbers where read is high
>>> compared to write
>>>
>>> 2. robocopy test:
>>> - mkfs.ext4 /dev/sda1
>>> - mount /dev/sda1 /media/disk
>>> - << configured samba >>
>>> - Mapped the /media/disk on windows
>>> - login on the mapped driver in windows
>>> - did a robocopy test, where write got 84MBps and read 14MBps
>>>
>>> read performance is too slow when compared to write in robocopy case.
>>> Can anyone help me out, how to debug this further.
>
> thanks!
> --
> Jagan.
On 14 November 2014 19:22, Roger Heflin <[email protected]> wrote:
> What kind of underlying disk is it?
GEN3 Sata link, SSD from Samsung.
>
> On Fri, Nov 14, 2014 at 7:36 AM, Jagan Teki <[email protected]> wrote:
>> On 14 November 2014 18:50, Roger Heflin <[email protected]> wrote:
>>> If you are robocoping small files you will hit other limits.
>>>
>>> Best I have seen with small files is around 30 files/second, and that
>>> involves multiple copies going on. Remember with a small files there
>>> are several reads and writes that need to be done to complete a create
>>> of a small file and each of these take time. 30 files/second ~ 30ms
>>> per file, not that bad considering that on a real spinning disk a
>>> single read/write op is 5-10ms, and creating the file entry, copying
>>> data and closing the file takes several operations (at least create
>>> file entry, write small amount of data, update file entry
>>> date/time/info). If the write in the middle is not a significant
>>> amount of data, the 2 extra ops are what hurts.
>>>
>>
>> But, I tried 4gb and 1gb files both got a similar numbers.
>>
>>> On Fri, Nov 14, 2014 at 6:55 AM, Jagan Teki <[email protected]> wrote:
>>>> Hi,
>>>>
>>>> I'm doing a performance testing on my bench ARM box.
>>>>
>>>> 1. dd test: I have validate the read and write by mounting /dev/sda1
>>>> with ext4 filesystem,
>>>> able to get the good performance numbers where read is high
>>>> compared to write
>>>>
>>>> 2. robocopy test:
>>>> - mkfs.ext4 /dev/sda1
>>>> - mount /dev/sda1 /media/disk
>>>> - << configured samba >>
>>>> - Mapped the /media/disk on windows
>>>> - login on the mapped driver in windows
>>>> - did a robocopy test, where write got 84MBps and read 14MBps
>>>>
>>>> read performance is too slow when compared to write in robocopy case.
>>>> Can anyone help me out, how to debug this further.
>>
>> thanks!
>> --
>> Jagan.
--
Jagan.