[RFC,0/9,v4,RESEND] Migrate Pages in lieu of discard

Message

Dave Hansen Oct. 7, 2020, 4:17 p.m. UTC

Changes since (automigrate-20200818):
 * Fall back to normal reclaim when demotion fails

The full series is also available here:

	https://github.com/hansendc/linux/tree/automigrate-20201007

I really just want folks to look at:

	[RFC][PATCH 5/9] mm/migrate: demote pages during reclaim

I've reworked that so that it can both use the high-level migration
API, and fall back to normal reclaim if migration fails.  I think
that gives us the best of both worlds.

I'm posting the series in case folks want to run the whole thing.

--

We're starting to see systems with more and more kinds of memory such
as Intel's implementation of persistent memory.

Let's say you have a system with some DRAM and some persistent memory.
Today, once DRAM fills up, reclaim will start and some of the DRAM
contents will be thrown out.  Allocations will, at some point, start
falling over to the slower persistent memory.

That has two nasty properties.  First, the newer allocations can end
up in the slower persistent memory.  Second, reclaimed data in DRAM
are just discarded even if there are gobs of space in persistent
memory that could be used.

This set implements a solution to these problems.  At the end of the
reclaim process in shrink_page_list() just before the last page
refcount is dropped, the page is migrated to persistent memory instead
of being dropped.

While I've talked about a DRAM/PMEM pairing, this approach would
function in any environment where memory tiers exist.

This is not perfect.  It "strands" pages in slower memory and never
brings them back to fast DRAM.  Other things need to be built to
promote hot pages back to DRAM.

This is also all based on an upstream mechanism that allows
persistent memory to be onlined and used as if it were volatile:

	http://lkml.kernel.org/r/[e-mail]

== Open Issues ==

 * For cpusets and memory policies that restrict allocations
   to PMEM, is it OK to demote to PMEM?  Do we need a cgroup-
   level API to opt-in or opt-out of these migrations?

--

Changes since (https://lwn.net/Articles/824830/):
 * Use higher-level migrate_pages() API approach from Yang Shi's
   earlier patches.
 * made sure to actually check node_reclaim_mode's new bit
 * disabled migration entirely before introducing RECLAIM_MIGRATE
 * Replace GFP_NOWAIT with explicit __GFP_KSWAPD_RECLAIM and
   comment why we want that.
 * Comment on effects of that keep multiple source nodes from
   sharing target nodes

Cc: Yang Shi <[e-mail]>
Cc: David Rientjes <[e-mail]>
Cc: Huang Ying <[e-mail]>
Cc: Dan Williams <[e-mail]>
Cc: David Hildenbrand <[e-mail]>

Comments

Yang Shi Oct. 12, 2020, 9:30 p.m. UTC | #1

On Wed, Oct 7, 2020 at 9:17 AM Dave Hansen <[e-mail]> wrote:
>
>
> Changes since (automigrate-20200818):
>  * Fall back to normal reclaim when demotion fails
>
> The full series is also available here:
>
>         https://github.com/hansendc/linux/tree/automigrate-20201007
>
> I really just want folks to look at:
>
>         [RFC][PATCH 5/9] mm/migrate: demote pages during reclaim
>
> I've reworked that so that it can both use the high-level migration
> API, and fall back to normal reclaim if migration fails.  I think
> that gives us the best of both worlds.

Thanks for doing this. Although I was inclined to think the kswapds on
PMEM nodes could make enough space for retrying migration later
instead of doing swap right away, this approach might be
over-engineering and over-killing. The simple immediate "retry regular
reclaim" approach also looks fine to me for the time being. We always
could optimize it later with more test results backed by real life
workloads.

>
> I'm posting the series in case folks want to run the whole thing.
>
> --
>
> We're starting to see systems with more and more kinds of memory such
> as Intel's implementation of persistent memory.
>
> Let's say you have a system with some DRAM and some persistent memory.
> Today, once DRAM fills up, reclaim will start and some of the DRAM
> contents will be thrown out.  Allocations will, at some point, start
> falling over to the slower persistent memory.
>
> That has two nasty properties.  First, the newer allocations can end
> up in the slower persistent memory.  Second, reclaimed data in DRAM
> are just discarded even if there are gobs of space in persistent
> memory that could be used.
>
> This set implements a solution to these problems.  At the end of the
> reclaim process in shrink_page_list() just before the last page
> refcount is dropped, the page is migrated to persistent memory instead
> of being dropped.
>
> While I've talked about a DRAM/PMEM pairing, this approach would
> function in any environment where memory tiers exist.
>
> This is not perfect.  It "strands" pages in slower memory and never
> brings them back to fast DRAM.  Other things need to be built to
> promote hot pages back to DRAM.
>
> This is also all based on an upstream mechanism that allows
> persistent memory to be onlined and used as if it were volatile:
>
>         http://lkml.kernel.org/r/[e-mail]
>
> == Open Issues ==
>
>  * For cpusets and memory policies that restrict allocations
>    to PMEM, is it OK to demote to PMEM?  Do we need a cgroup-
>    level API to opt-in or opt-out of these migrations?
>
> --
>
> Changes since (https://lwn.net/Articles/824830/):
>  * Use higher-level migrate_pages() API approach from Yang Shi's
>    earlier patches.
>  * made sure to actually check node_reclaim_mode's new bit
>  * disabled migration entirely before introducing RECLAIM_MIGRATE
>  * Replace GFP_NOWAIT with explicit __GFP_KSWAPD_RECLAIM and
>    comment why we want that.
>  * Comment on effects of that keep multiple source nodes from
>    sharing target nodes
>
> Cc: Yang Shi <[e-mail]>
> Cc: David Rientjes <[e-mail]>
> Cc: Huang Ying <[e-mail]>
> Cc: Dan Williams <[e-mail]>
> Cc: David Hildenbrand <[e-mail]>
>