Re: [PATCH 0/13] Parallel struct page initialisation v4

From: Waiman Long
Date: Tue Apr 28 2015 - 21:16:12 EST


On 04/28/2015 10:36 AM, Mel Gorman wrote:
The bulk of the changes here are related to Andrew's feedback. Functionally
there is almost no difference.

Changelog since v3
o Fix section-related warning
o Comments, clarifications, checkpatch
o Report the number of pages initialised

Changelog since v2
o Reduce overhead of topology_init
o Remove boot-time kernel parameter to enable/disable
o Enable on UMA

Changelog since v1
o Always initialise low zones
o Typo corrections
o Rename parallel mem init to parallel struct page init
o Rebase to 4.0

Struct page initialisation had been identified as one of the reasons why
large machines take a long time to boot. Patches were posted a long time ago
to defer initialisation until they were first used. This was rejected on
the grounds it should not be necessary to hurt the fast paths. This series
reuses much of the work from that time but defers the initialisation of
memory to kswapd so that one thread per node initialises memory local to
that node.

After applying the series and setting the appropriate Kconfig variable I
see this in the boot log on a 64G machine

[ 7.383764] kswapd 0 initialised deferred memory in 188ms
[ 7.404253] kswapd 1 initialised deferred memory in 208ms
[ 7.411044] kswapd 3 initialised deferred memory in 216ms
[ 7.411551] kswapd 2 initialised deferred memory in 216ms

On a 1TB machine, I see

[ 8.406511] kswapd 3 initialised deferred memory in 1116ms
[ 8.428518] kswapd 1 initialised deferred memory in 1140ms
[ 8.435977] kswapd 0 initialised deferred memory in 1148ms
[ 8.437416] kswapd 2 initialised deferred memory in 1148ms

Once booted the machine appears to work as normal. Boot times were measured
from the time shutdown was called until ssh was available again. In the
64G case, the boot time savings are negligible. On the 1TB machine, the
savings were 16 seconds.

It would be nice if the people that have access to really large machines
would test this series and report how much boot time is reduced.



I ran a bootup timing test on a 12-TB 16-socket IvyBridge-EX system. From grub menu to ssh login, the bootup time was 453s before the patch and 265s after the patch - a saving of 188s (42%). I used a different OS environment and config file with this test and so the timing data weren't comparable with my previous testing data. The kswapd log entries were

[ 45.973967] kswapd 4 initialised 197655470 pages in 4390ms
[ 45.974214] kswapd 7 initialised 197655470 pages in 4390ms
[ 45.976692] kswapd 15 initialised 197654299 pages in 4390ms
[ 45.993284] kswapd 0 initialised 197131131 pages in 4410ms
[ 46.032735] kswapd 9 initialised 197655470 pages in 4447ms
[ 46.065856] kswapd 8 initialised 197655470 pages in 4481ms
[ 46.066615] kswapd 1 initialised 197622702 pages in 4483ms
[ 46.077995] kswapd 2 initialised 197655470 pages in 4495ms
[ 46.219508] kswapd 13 initialised 197655470 pages in 4633ms
[ 46.224358] kswapd 3 initialised 197655470 pages in 4641ms
[ 46.228441] kswapd 11 initialised 197655470 pages in 4643ms
[ 46.232258] kswapd 12 initialised 197655470 pages in 4647ms
[ 46.239659] kswapd 10 initialised 197655470 pages in 4654ms
[ 46.243402] kswapd 14 initialised 197655470 pages in 4657ms
[ 46.250368] kswapd 5 initialised 197655470 pages in 4666ms
[ 46.254659] kswapd 6 initialised 197655470 pages in 4670ms

Cheers,
Longman
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/