I tested this [1] patch on top of 5.16-rc2, the same tests. - with noswap: 2021-12-02 19:41:19,279: Stall times for the last 146.5s: 2021-12-02 19:41:19,279: ----------- 2021-12-02 19:41:19,279: some cpu 1.1s, avg 0.8% 2021-12-02 19:41:19,279: ----------- 2021-12-02 19:41:19,279: some io 116.2s, avg 79.3% 2021-12-02 19:41:19,280: full io 109.6s, avg 74.8% 2021-12-02 19:41:19,280: ----------- 2021-12-02 19:41:19,280: some memory 3.9s, avg 2.6% 2021-12-02 19:41:19,280: full memory 3.8s, avg 2.6% Excellent! psi log: https://raw.githubusercontent.com/hakavlad/cache-tests/main/516-reclaim-throttle/516-rc2/patch6/noswap/psi mem log: https://raw.githubusercontent.com/hakavlad/cache-tests/main/516-reclaim-throttle/516-rc2/patch6/noswap/mem - with swappiness=0 2021-12-02 19:46:04,860: Stall times for the last 144.5s: 2021-12-02 19:46:04,860: ----------- 2021-12-02 19:46:04,860: some cpu 1.1s, avg 0.8% 2021-12-02 19:46:04,860: ----------- 2021-12-02 19:46:04,860: some io 106.9s, avg 74.0% 2021-12-02 19:46:04,861: full io 101.3s, avg 70.1% 2021-12-02 19:46:04,861: ----------- 2021-12-02 19:46:04,861: some memory 99.6s, avg 68.9% 2021-12-02 19:46:04,861: full memory 95.6s, avg 66.2% PSI mem pressure was high, but there were no long stalls. psi log: https://raw.githubusercontent.com/hakavlad/cache-tests/main/516-reclaim-throttle/516-rc2/patch6/swappiness0/psi mem log: https://raw.githubusercontent.com/hakavlad/cache-tests/main/516-reclaim-throttle/516-rc2/patch6/swappiness0/mem [1] https://lore.kernel.org/all/20211201140005.GU3366@xxxxxxxxxxxxxxxxxxx/
