Re: [RFC PATCH v2 0/7] DAMON based 2-tier memory management for CXL memory

From: Honggyu Kim
Date: Fri Mar 22 2024 - 05:03:12 EST


Hi SeongJae,

On Tue, 27 Feb 2024 15:51:20 -0800 SeongJae Park <sj@xxxxxxxxxx> wrote:
> On Mon, 26 Feb 2024 23:05:46 +0900 Honggyu Kim <honggyu.kim@xxxxxx> wrote:
>
> > There was an RFC IDEA "DAMOS-based Tiered-Memory Management" previously
> > posted at [1].
> >
> > It says there is no implementation of the demote/promote DAMOS action
> > are made. This RFC is about its implementation for physical address
> > space.
> >
> >
> > Introduction
> > ============
> >
> > With the advent of CXL/PCIe attached DRAM, which will be called simply
> > as CXL memory in this cover letter, some systems are becoming more
> > heterogeneous having memory systems with different latency and bandwidth
> > characteristics. They are usually handled as different NUMA nodes in
> > separate memory tiers and CXL memory is used as slow tiers because of
> > its protocol overhead compared to local DRAM.
> >
> > In this kind of systems, we need to be careful placing memory pages on
> > proper NUMA nodes based on the memory access frequency. Otherwise, some
> > frequently accessed pages might reside on slow tiers and it makes
> > performance degradation unexpectedly. Moreover, the memory access
> > patterns can be changed at runtime.
> >
> > To handle this problem, we need a way to monitor the memory access
> > patterns and migrate pages based on their access temperature. The
> > DAMON(Data Access MONitor) framework and its DAMOS(DAMON-based Operation
> > Schemes) can be useful features for monitoring and migrating pages.
> > DAMOS provides multiple actions based on DAMON monitoring results and it
> > can be used for proactive reclaim, which means swapping cold pages out
> > with DAMOS_PAGEOUT action, but it doesn't support migration actions such
> > as demotion and promotion between tiered memory nodes.
> >
> > This series supports two new DAMOS actions; DAMOS_DEMOTE for demotion
> > from fast tiers and DAMOS_PROMOTE for promotion from slow tiers. This
> > prevents hot pages from being stuck on slow tiers, which makes
> > performance degradation and cold pages can be proactively demoted to
> > slow tiers so that the system can increase the chance to allocate more
> > hot pages to fast tiers.
> >
> > The DAMON provides various tuning knobs but we found that the proactive
> > demotion for cold pages is especially useful when the system is running
> > out of memory on its fast tier nodes.
> >
> > Our evaluation result shows that it reduces the performance slowdown
> > compared to the default memory policy from 15~17% to 4~5% when the
> > system runs under high memory pressure on its fast tier DRAM nodes.
> >
> >
> > DAMON configuration
> > ===================
> >
> > The specific DAMON configuration doesn't have to be in the scope of this
> > patch series, but some rough idea is better to be shared to explain the
> > evaluation result.
> >
> > The DAMON provides many knobs for fine tuning but its configuration file
> > is generated by HMSDK[2]. It includes gen_config.py script that
> > generates a json file with the full config of DAMON knobs and it creates
> > multiple kdamonds for each NUMA node when the DAMON is enabled so that
> > it can run hot/cold based migration for tiered memory.
>
> I was feeling a bit confused from here since DAMON doesn't receive parameters
> via a file. To my understanding, the 'configuration file' means the input file
> for DAMON user-space tool, damo, not DAMON. Just a trivial thing, but making
> it clear if possible could help readers in my opinion.
>
> >
> >
> > Evaluation Workload
> > ===================
> >
> > The performance evaluation is done with redis[3], which is a widely used
> > in-memory database and the memory access patterns are generated via
> > YCSB[4]. We have measured two different workloads with zipfian and
> > latest distributions but their configs are slightly modified to make
> > memory usage higher and execution time longer for better evaluation.
> >
> > The idea of evaluation using these demote and promote actions covers
> > system-wide memory management rather than partitioning hot/cold pages of
> > a single workload. The default memory allocation policy creates pages
> > to the fast tier DRAM node first, then allocates newly created pages to
> > the slow tier CXL node when the DRAM node has insufficient free space.
> > Once the page allocation is done then those pages never move between
> > NUMA nodes. It's not true when using numa balancing, but it is not the
> > scope of this DAMON based 2-tier memory management support.
> >
> > If the working set of redis can be fit fully into the DRAM node, then
> > the redis will access the fast DRAM only. Since the performance of DRAM
> > only is faster than partially accessing CXL memory in slow tiers, this
> > environment is not useful to evaluate this patch series.
> >
> > To make pages of redis be distributed across fast DRAM node and slow
> > CXL node to evaluate our demote and promote actions, we pre-allocate
> > some cold memory externally using mmap and memset before launching
> > redis-server. We assumed that there are enough amount of cold memory in
> > datacenters as TMO[5] and TPP[6] papers mentioned.
> >
> > The evaluation sequence is as follows.
> >
> > 1. Turn on DAMON with DAMOS_DEMOTE action for DRAM node and
> > DAMOS_PROMOTE action for CXL node. It demotes cold pages on DRAM
> > node and promotes hot pages on CXL node in a regular interval.
> > 2. Allocate a huge block of cold memory by calling mmap and memset at
> > the fast tier DRAM node, then make the process sleep to make the fast
> > tier has insufficient memory for redis-server.
> > 3. Launch redis-server and load prebaked snapshot image, dump.rdb. The
> > redis-server consumes 52GB of anon pages and 33GB of file pages, but
> > due to the cold memory allocated at 2, it fails allocating the entire
> > memory of redis-server on the fast tier DRAM node so it partially
> > allocates the remaining on the slow tier CXL node. The ratio of
> > DRAM:CXL depends on the size of the pre-allocated cold memory.
> > 4. Run YCSB to make zipfian or latest distribution of memory accesses to
> > redis-server, then measure its execution time when it's completed.
> > 5. Repeat 4 over 50 times to measure the average execution time for each
> > run.
> > 6. Increase the cold memory size then repeat goes to 2.
> >
> > For each test at 4 took about a minute so repeating it 50 times almost
> > took about 1 hour for each test with a specific cold memory from 440GB
> > to 500GB in 10GB increments for each evaluation. So it took about more
> > than 10 hours for both zipfian and latest workloads to get the entire
> > evaluation results. Repeating the same test set multiple times doesn't
> > show much difference so I think it might be enough to make the result
> > reliable.
> >
> >
> > Evaluation Results
> > ==================
> >
> > All the result values are normalized to DRAM-only execution time because
> > the workload cannot be faster than DRAM-only unless the workload hits
> > the bandwidth peak but our redis test doesn't go beyond the bandwidth
> > limit.
> >
> > So the DRAM-only execution time is the ideal result without affected by
> > the gap between DRAM and CXL performance difference. The NUMA node
> > environment is as follows.
> >
> > node0 - local DRAM, 512GB with a CPU socket (fast tier)
> > node1 - disabled
> > node2 - CXL DRAM, 96GB, no CPU attached (slow tier)
> >
> > The following is the result of generating zipfian distribution to
> > redis-server and the numbers are averaged by 50 times of execution.
> >
> > 1. YCSB zipfian distribution read only workload
> > memory pressure with cold memory on node0 with 512GB of local DRAM.
> > =============+================================================+=========
> > | cold memory occupied by mmap and memset |
> > | 0G 440G 450G 460G 470G 480G 490G 500G |
> > =============+================================================+=========
> > Execution time normalized to DRAM-only values | GEOMEAN
> > -------------+------------------------------------------------+---------
> > DRAM-only | 1.00 - - - - - - - | 1.00
> > CXL-only | 1.21 - - - - - - - | 1.21
> > default | - 1.09 1.10 1.13 1.15 1.18 1.21 1.21 | 1.15
> > DAMON 2-tier | - 1.02 1.04 1.05 1.04 1.05 1.05 1.06 | 1.04
> > =============+================================================+=========
> > CXL usage of redis-server in GB | AVERAGE
> > -------------+------------------------------------------------+---------
> > DRAM-only | 0.0 - - - - - - - | 0.0
> > CXL-only | 52.6 - - - - - - - | 52.6
> > default | - 19.4 26.1 32.3 38.5 44.7 50.5 50.3 | 37.4
> > DAMON 2-tier | - 0.1 1.6 5.2 8.0 9.1 11.8 13.6 | 7.1
> > =============+================================================+=========
> >
> > Each test result is based on the exeuction environment as follows.
> >
> > DRAM-only : redis-server uses only local DRAM memory.
> > CXL-only : redis-server uses only CXL memory.
> > default : default memory policy(MPOL_DEFAULT).
> > numa balancing disabled.
> > DAMON 2-tier: DAMON enabled with DAMOS_DEMOTE for DRAM nodes and
> > DAMOS_PROMOTE for CXL nodes.
> >
> > The above result shows the "default" execution time goes up as the size
> > of cold memory is increased from 440G to 500G because the more cold
> > memory used, the more CXL memory is used for the target redis workload
> > and this makes the execution time increase.
> >
> > However, "DAMON 2-tier" result shows less slowdown because the
> > DAMOS_DEMOTE action at DRAM node proactively demotes pre-allocated cold
> > memory to CXL node and this free space at DRAM increases more chance to
> > allocate hot or warm pages of redis-server to fast DRAM node. Moreover,
> > DEMOS_PROMOTE action at CXL node also promotes hot pages of redis-server
> > to DRAM node actively.
> >
> > As a result, it makes more memory of redis-server stay in DRAM node
> > compared to "default" memory policy and this makes the performance
> > improvement.
> >
> > The following result of latest distribution workload shows similar data.
> >
> > 2. YCSB latest distribution read only workload
> > memory pressure with cold memory on node0 with 512GB of local DRAM.
> > =============+================================================+=========
> > | cold memory occupied by mmap and memset |
> > | 0G 440G 450G 460G 470G 480G 490G 500G |
> > =============+================================================+=========
> > Execution time normalized to DRAM-only values | GEOMEAN
> > -------------+------------------------------------------------+---------
> > DRAM-only | 1.00 - - - - - - - | 1.00
> > CXL-only | 1.18 - - - - - - - | 1.18
> > default | - 1.16 1.15 1.17 1.18 1.16 1.18 1.15 | 1.17
> > DAMON 2-tier | - 1.04 1.04 1.05 1.05 1.06 1.05 1.06 | 1.05
> > =============+================================================+=========
> > CXL usage of redis-server in GB | AVERAGE
> > -------------+------------------------------------------------+---------
> > DRAM-only | 0.0 - - - - - - - | 0.0
> > CXL-only | 52.6 - - - - - - - | 52.6
> > default | - 19.3 26.1 32.2 38.5 44.6 50.5 50.6 | 37.4
> > DAMON 2-tier | - 1.3 3.8 7.0 4.1 9.4 12.5 16.7 | 7.8
> > =============+================================================+=========
> >
> > In summary of both results, our evaluation shows that "DAMON 2-tier"
> > memory management reduces the performance slowdown compared to the
> > "default" memory policy from 15~17% to 4~5% when the system runs with
> > high memory pressure on its fast tier DRAM nodes.
> >
> > The similar evaluation was done in another machine that has 256GB of
> > local DRAM and 96GB of CXL memory. The performance slowdown is reduced
> > from 20~24% for "default" to 5~7% for "DAMON 2-tier".
> >
> > Having these DAMOS_DEMOTE and DAMOS_PROMOTE actions can make 2-tier
> > memory systems run more efficiently under high memory pressures.
>
> Thank you for running the tests again with the new version of the patches and
> sharing the results!

It's a bit late answer, but the result was from the previous evaluation.
I ran it again with RFC v2, but didn't see much difference so just
pasted the same result here.

> >
> > Signed-off-by: Honggyu Kim <honggyu.kim@xxxxxx>
> > Signed-off-by: Hyeongtak Ji <hyeongtak.ji@xxxxxx>
> > Signed-off-by: Rakie Kim <rakie.kim@xxxxxx>
> >
> > [1] https://lore.kernel.org/damon/20231112195602.61525-1-sj@xxxxxxxxxx
> > [2] https://github.com/skhynix/hmsdk
> > [3] https://github.com/redis/redis/tree/7.0.0
> > [4] https://github.com/brianfrankcooper/YCSB/tree/0.17.0
> > [5] https://dl.acm.org/doi/10.1145/3503222.3507731
> > [6] https://dl.acm.org/doi/10.1145/3582016.3582063
> >
> > Changes from RFC:
> > 1. Move most of implementation from mm/vmscan.c to mm/damon/paddr.c.
> > 2. Simplify some functions of vmscan.c and used in paddr.c, but need
> > to be reviewed more in depth.
> > 3. Refactor most functions for common usage for both promote and
> > demote actions and introduce an enum migration_mode for its control.
> > 4. Add "target_nid" sysfs knob for migration destination node for both
> > promote and demote actions.
> > 5. Move DAMOS_PROMOTE before DAMOS_DEMOTE and move then even above
> > DAMOS_STAT.
>
> Thank you very much for addressing many of my comments.

Thanks for your feedback in details.

> >
> > Honggyu Kim (3):
> > mm/damon: refactor DAMOS_PAGEOUT with migration_mode
> > mm: make alloc_demote_folio externally invokable for migration
> > mm/damon: introduce DAMOS_DEMOTE action for demotion
> >
> > Hyeongtak Ji (4):
> > mm/memory-tiers: add next_promotion_node to find promotion target
> > mm/damon: introduce DAMOS_PROMOTE action for promotion
> > mm/damon/sysfs-schemes: add target_nid on sysfs-schemes
> > mm/damon/sysfs-schemes: apply target_nid for promote and demote
> > actions
>
> Honggyu joined DAMON Beer/Coffee/Tea Chat[1] yesterday, and we discussed about
> this patchset in high level. Sharing the summary here for open discussion. As
> also discussed on the first version of this patchset[2], we want to make single
> action for general page migration with minimum changes, but would like to keep
> page level access re-check. We also agreed the previously proposed DAMOS
> filter-based approach could make sense for the purpose.

Thanks very much for the summary. I have been trying to merge promote
and demote actions into a single migrate action, but I found an issue
regarding damon_pa_scheme_score. It currently calls damon_cold_score()
for demote action and damon_hot_score() for promote action, but what
should we call when we use a single migrate action?

Thanks,
Honggyu

> Because I was anyway planning making such DAMOS filter for not only
> promotion/demotion but other types of DAMOS action, I will start developing the
> page level access re-check results based DAMOS filter. Once the implementation
> of the prototype is done, I will share the early implementation. Then, Honggyu
> will adjust their implementation based on the filter, and run their tests again
> and share the results.
>
> [1] https://lore.kernel.org/damon/20220810225102.124459-1-sj@xxxxxxxxxx/
> [2] https://lore.kernel.org/damon/20240118171756.80356-1-sj@xxxxxxxxxx
>
>
> Thanks,
> SJ
>
> >
> > include/linux/damon.h | 15 +-
> > include/linux/memory-tiers.h | 11 ++
> > include/linux/migrate_mode.h | 1 +
> > include/linux/vm_event_item.h | 1 +
> > include/trace/events/migrate.h | 3 +-
> > mm/damon/core.c | 5 +-
> > mm/damon/dbgfs.c | 2 +-
> > mm/damon/lru_sort.c | 3 +-
> > mm/damon/paddr.c | 282 ++++++++++++++++++++++++++++++++-
> > mm/damon/reclaim.c | 3 +-
> > mm/damon/sysfs-schemes.c | 39 ++++-
> > mm/internal.h | 1 +
> > mm/memory-tiers.c | 43 +++++
> > mm/vmscan.c | 10 +-
> > mm/vmstat.c | 1 +
> > 15 files changed, 404 insertions(+), 16 deletions(-)
> >
> >
> > base-commit: 0dd3ee31125508cd67f7e7172247f05b7fd1753a
> > --
> > 2.34.1