LIO performance Issue

LIO performance Issue

[Abhilash Sankar - ERS, HCL Tech]

Hi all,

My name is Abhilash, I am new to this community, I am trying to do a performance study of LIO-iSCSI and SCST using robocopy (I/O Tool).
I have successfully ported LIO to my Debian OS. Based on my initial analysis, SCST had shown a better write performance than LIO.

Read performance seems to be almost similar for both the stack. But write performance for file based LIO-iSCSI has degraded more than 50%
while compared to SCST stack in my setup.

I am using LIO 4.0, file based IO and Windows 7 initiator.

It would be much helpful if anyone could give me some suggestion to improve the LIO write performance.

Regards,
Abhilash


::DISCLAIMER::
-----------------------------------------------------------------------------------------------------------------------

The contents of this e-mail and any attachment(s) are confidential and intended for the named recipient(s) only.
It shall not attach any liability on the originator or HCL or its affiliates. Any views or opinions presented in
this email are solely those of the author and may not necessarily reflect the opinions of HCL or its affiliates.
Any form of reproduction, dissemination, copying, disclosure, modification, distribution and / or publication of
this message without the prior written consent of the author of this e-mail is strictly prohibited. If you have
received this email in error please delete it and notify the sender immediately. Before opening any mail and
attachments please check them for viruses and defect.

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

Re: LIO performance Issue

[Nicholas A. Bellinger]

On Wed, 2011-09-14 at 19:30 +0530, Abhilash Sankar - ERS, HCL Tech
wrote:
> Hi all,
> 
> My name is Abhilash, I am new to this community, I am trying to do a performance study of LIO-iSCSI and SCST using robocopy (I/O Tool).
> I have successfully ported LIO to my Debian OS. Based on my initial analysis, SCST had shown a better write performance than LIO.
> 
> Read performance seems to be almost similar for both the stack. But write performance for file based LIO-iSCSI has degraded more than 50%
> while compared to SCST stack in my setup.
> 
> I am using LIO 4.0, file based IO and Windows 7 initiator.
> 
> It would be much helpful if anyone could give me some suggestion to improve the LIO write performance.
> 

Hi Abhilash,

By default target_core_file.c is using O_SYNC to ensure that WRITEs are
written directly to disk before acknowledging back to the SCSI client.
This will have a considerable effect on FILEIO WRITE performance, but is
done by default to ensure proper data integrity in the face of target
power loss.

You may want to double check the filp_open() flags with SCST to ensure
that you are comparing 'apples to apples' here.  Optionally you can try
disabling O_SYNC with target_core_file.c via rtslib with the class
FileIOStorageObject:_set_buffered_mode() mentioned here:

http://www.risingtidesystems.com/git/?p=rtslib.git;a=blob;f=rtslib/tcm.py;h=4da49b0f4a4aea9a1eebed23d457181085a2d03d;hb=HEAD#l975

--nab

Re: LIO performance Issue

[Ravi Shankar]

>>     
>
> Hi Abhilash,
>
> By default target_core_file.c is using O_SYNC to ensure that WRITEs are
> written directly to disk before acknowledging back to the SCSI client.
> This will have a considerable effect on FILEIO WRITE performance, but is
> done by default to ensure proper data integrity in the face of target
> power loss.
>
> You may want to double check the filp_open() flags with SCST to ensure
> that you are comparing 'apples to apples' here.  Optionally you can try
> disabling O_SYNC with target_core_file.c via rtslib with the class
> FileIOStorageObject:_set_buffered_mode() mentioned here:
>
>   
Another area to consider improvement is how SCSI Target Data phase are 
handled.
The target framework should split SCSI data phases into multiple smaller 
IO to increase
performance. In other work instead of synchronous data phase it need to 
be asynchronous.

--ravi

Re: LIO performance Issue

[Christoph Hellwig]

On Thu, Sep 15, 2011 at 12:47:26PM -0700, Ravi Shankar wrote:
> Another area to consider improvement is how SCSI Target Data phase
> are handled.
> The target framework should split SCSI data phases into multiple
> smaller IO to increase
> performance. In other work instead of synchronous data phase it need
> to be asynchronous.

There isn't any good in-kernel interface to do asynchronous file I/O.
If you use the iblock backend you'll get asynchronous behaviour.