There are currently several issues with zswap writeback: 1. There is only a single global LRU for zswap. This makes it impossible to perform worload-specific shrinking - an memcg under memory pressure cannot determine which pages in the pool it owns, and often ends up writing pages from other memcgs. This issue has been previously observed in practice and mitigated by simply disabling memcg-initiated shrinking: https://lore.kernel.org/all/20230530232435.3097106-1-nphamcs@xxxxxxxxx/T/#u But this solution leaves a lot to be desired, as we still do not have an avenue for an memcg to free up its own memory locked up in zswap. 2. We only shrink the zswap pool when the user-defined limit is hit. This means that if we set the limit too high, cold data that are unlikely to be used again will reside in the pool, wasting precious memory. It is hard to predict how much zswap space will be needed ahead of time, as this depends on the workload (specifically, on factors such as memory access patterns and compressibility of the memory pages). This patch series solves these issues by separating the global zswap LRU into per-memcg and per-NUMA LRUs, and performs workload-specific (i.e memcg- and NUMA-aware) zswap writeback under memory pressure. The new shrinker does not have any parameter that must be tuned by the user, and can be opted in or out on a per-memcg basis. On a benchmark that we have run: (without the shrinker) real -- mean: 153.27s, median: 153.199s sys -- mean: 541.652s, median: 541.903s user -- mean: 4384.9673999999995s, median: 4385.471s (with the shrinker) real -- mean: 151.4956s, median: 151.456s sys -- mean: 461.14639999999997s, median: 465.656s user -- mean: 4384.7118s, median: 4384.675s We observed a 14-15% reduction in kernel CPU time, which translated to over 1% reduction in real time. On another benchmark, where there was a lot more cold memory residing in zswap, we observed even more pronounced gains: (without the shrinker) real -- mean: 157.52519999999998s, median: 157.281s sys -- mean: 769.3082s, median: 780.545s user -- mean: 4378.1622s, median: 4378.286s (with the shrinker) real -- mean: 152.9608s, median: 152.845s sys -- mean: 517.4446s, median: 506.749s user -- mean: 4387.694s, median: 4387.935s Here, we saw around 32-35% reduction in kernel CPU time, which translated to 2.8% reduction in real time. These results confirm our hypothesis that the shrinker is more helpful the more cold memory we have. Domenico Cerasuolo (1): zswap: make shrinking memcg-aware Nhat Pham (1): zswap: shrinks zswap pool based on memory pressure Documentation/admin-guide/mm/zswap.rst | 12 + include/linux/list_lru.h | 39 +++ include/linux/memcontrol.h | 1 + include/linux/mmzone.h | 14 + include/linux/zswap.h | 9 + mm/list_lru.c | 46 ++- mm/memcontrol.c | 33 +++ mm/swap_state.c | 50 +++- mm/zswap.c | 369 ++++++++++++++++++++++--- 9 files changed, 518 insertions(+), 55 deletions(-) -- 2.34.1
