[PATCH 0/8] Avoiding fragmentation with subzone groupings v25

[Mel Gorman <mel@csn.ul.ie>]

This is the latest version of anti-fragmentation based on sub-zones (previously
called list-based anti-fragmentation) based on top of 2.6.18-rc5-mm1. In
it's last release, it was decided that the scheme should be implemented with
zones to avoid affecting the page allocator hot paths. However, at VM Summit,
it was made clear that zones may not be the right answer either because zones
have their own issues. Hence, this is a reintroduction of the first approach.

The purpose of these patches is to reduce external fragmentation by grouping
pages of related types together. The objective is that when page reclaim
occurs, there is a greater chance that large contiguous pages will be
free. Note that this is not a defragmentation which would get contiguous
pages by moving pages around.

This patch works by categorising allocations by their reclaimability;

EasyReclaimable - These are userspace pages that are easily reclaimable. This
	flag is set when it is known that the pages will be trivially reclaimed
	by writing the page out to swap or syncing with backing storage

KernelReclaimable - These are allocations for some kernel caches that are
	reclaimable or allocations that are known to be very short-lived.

KernelNonReclaimable - These are pages that are allocated by the kernel that
	are not trivially reclaimed. For example, the memory allocated for a
	loaded module would be in this category. By default, allocations are
	considered to be of this type

Instead of having one MAX_ORDER-sized array of free lists in struct free_area,
there is one for each type of reclaimability. Once a 2^MAX_ORDER block of
pages is split for a type of allocation, it is added to the free-lists for
that type, in effect reserving it. Hence, over time, pages of the different
types can be clustered together. When a page is allocated, the page-flags
are updated with a value indicating it's type of reclaimability so that it
is placed on the correct list on free.

When the preferred freelists are expired, the largest possible block is taken
from an alternative list. Buddies that are split from that large block are
placed on the preferred allocation-type freelists to mitigate fragmentation.

This implementation gives best-effort for low fragmentation in all zones. To
be effective, min_free_kbytes needs to be set to a value about 10% of physical
memory (10% was found by experimentation, it may be workload dependant). To get
that value lower, anti-fragmentation needs to be significantly more invasive
so it's best to find out what sorts of workloads still cause fragmentation
before taking further steps.

Our tests show that about 60-70% of physical memory can be allocated on
a desktop after a few days uptime. In benchmarks and stress tests, we are
finding that 80% of memory is available as contiguous blocks at the end of
the test. To compare, a standard kernel was getting < 1% of memory as large
pages on a desktop and about 8-12% of memory as large pages at the end of
stress tests.

Performance tests are within 0.1% for kbuild on a number of test machines. aim9
is usually within 1% except on x86_64 where aim9 results are unreliable.
I have never been able to show it but it is possible the main allocator path
is adversely affected by anti-fragmentation and it may be exposed by using
differnet compilers or benchmarks. If any regressions are detected due to
anti-fragmentation, it may be simply disabled via the kernel configuration
and I'd appreciate a report detailing the regression and how to trigger it.

Following this email are 8 patches.  They early patches introduce the split
between user and kernel allocations.  Later we introduce a further split
for kernel allocations, into KernRclm and KernNoRclm.  Note that although
in early patches an additional page flag is consumed, later patches reuse
the suspend bits, releasing this bit again.

Comments?
-- 
-- 
Mel Gorman
Part-time Phd Student                          Linux Technology Center
University of Limerick                         IBM Dublin Software Lab
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/