swap behavior during fast allocation

swap behavior during fast allocation

[Luigi Semenzato]

Greetings MM experts,

we (Chrome OS) are experiencing episodes of extremely sluggish
interactive response, i.e. tens of seconds delay between an action
(such as typing something) and the corresponding screen update during
times of very fast allocation, while using zram.

We can reproduce this by running a few processes that mmap large
chunks of memory, then randomly touch pages to fault them in.  We also
think this happens when a process writes a large amount of data using
buffered I/O, and the "Buffers" field in /proc/meminfo exceeds 1GB.
(This is something that can and should be corrected by using
unbuffered I/O instead, but it's a data point.)

We're wondering if this problem has been noticed before and what folks
do to ameliorate the situation.  Specifically, is there any way to
throttle the rate of allocation when the system is swapping?

Thanks!
Luigi

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: swap behavior during fast allocation

[Dave Hansen]

On 08/05/2013 11:04 AM, Luigi Semenzato wrote:
> We can reproduce this by running a few processes that mmap large
> chunks of memory, then randomly touch pages to fault them in.  We also
> think this happens when a process writes a large amount of data using
> buffered I/O, and the "Buffers" field in /proc/meminfo exceeds 1GB.
> (This is something that can and should be corrected by using
> unbuffered I/O instead, but it's a data point.)

Where are all the buffers coming from?  Most I/O to/from filesystems
should be instantiating relatively modest amounts of Buffers.  Are you
doing I/O directly to devices for some reason?

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: swap behavior during fast allocation

[Luigi Semenzato]

On Mon, Aug 5, 2013 at 11:24 AM, Dave Hansen <dave.hansen@intel.com> wrote:
> On 08/05/2013 11:04 AM, Luigi Semenzato wrote:
>> We can reproduce this by running a few processes that mmap large
>> chunks of memory, then randomly touch pages to fault them in.  We also
>> think this happens when a process writes a large amount of data using
>> buffered I/O, and the "Buffers" field in /proc/meminfo exceeds 1GB.
>> (This is something that can and should be corrected by using
>> unbuffered I/O instead, but it's a data point.)
>
> Where are all the buffers coming from?  Most I/O to/from filesystems
> should be instantiating relatively modest amounts of Buffers.  Are you
> doing I/O directly to devices for some reason?

Correct.  This is the autoupdate process, which writes the changed
kernel and filesystem blocks directly to raw partitions.  (The
filesystem partition is obviously not currently in use.)

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>