Re: [Patch 2/8] Block I/O, swapin delays

[Shailabh Nagar <nagar@watson.ibm.com>]

Andrew Morton wrote:

>Shailabh Nagar <nagar@watson.ibm.com> wrote:
>  
>
>>delayacct-blkio-swapin.patch
>>
>>Collect per-task block I/O delay statistics.
>>
>>Unlike earlier iterations of the delay accounting
>>patches, now delays are only collected for the actual
>>I/O waits rather than try and cover the delays seen in
>>I/O submission paths.
>>
>>Account separately for block I/O delays
>>incurred as a result of swapin page faults whose
>>frequency can be affected by the task/process' rss limit.
>>Hence swapin delays can act as feedback for rss limit changes
>>independent of I/O priority changes.
>>
>>..
>>
>>+#define PF_SWAPIN	0x02000000	/* I am doing a swap in */
>>
>>    
>>
>
>Is there really no sane alternative to doing it this way?
>  
>

I'm not sure there is. This goes back to what is really being measured 
as "block I/O delay".

Earlier iterations of the code tried to measure total block I/O delay 
including I/O submission
and the wait for completion:
    http://www.uwsg.indiana.edu/hypermail/linux/kernel/0602.3/0905.html
In that design, the swapin delay was being measured as time spent in the 
do_swap_page
function:
    http://www.uwsg.indiana.edu/hypermail/linux/kernel/0602.3/0906.html

However, Arjan pointed out the difficulties in trying to account for all 
I/O submission paths
so we reverted to measuring just the block I/O wait time within 
io_schedule/io_schedule_timeout.

Since these functions will be called as a result of do_swap_page, the 
only way to distinguish a
block I/O being done for a swapin vs. others was to mark the task struct 
in some way and
task->flags seemed like a convenient way to do it.

We could easily introduce and use a separate flag variable in the delay 
accounting structure itself if
using up another bit of task->flags is a concern ?

Or do you have doubts about the methodology of using a flag itself ? 
Don't see any other lightweight
way of doing this.

--Shailabh