1 .. SPDX-License-Identifier: GPL-2.0
2
3 =============
4 Multi-Gen LRU
5 =============
6 The multi-gen LRU is an alternative LRU implem
7 page reclaim and improves performance under me
8 reclaim decides the kernel's caching policy an
9 memory. It directly impacts the kswapd CPU usa
10
11 Quick start
12 ===========
13 Build the kernel with the following configurat
14
15 * ``CONFIG_LRU_GEN=y``
16 * ``CONFIG_LRU_GEN_ENABLED=y``
17
18 All set!
19
20 Runtime options
21 ===============
22 ``/sys/kernel/mm/lru_gen/`` contains stable AB
23 following subsections.
24
25 Kill switch
26 -----------
27 ``enabled`` accepts different values to enable
28 following components. Its default value depend
29 ``CONFIG_LRU_GEN_ENABLED``. All the components
30 unless some of them have unforeseen side effec
31 ``enabled`` has no effect when a component is
32 hardware, and valid values will be accepted ev
33 is off.
34
35 ====== =======================================
36 Values Components
37 ====== =======================================
38 0x0001 The main switch for the multi-gen LRU.
39 0x0002 Clearing the accessed bit in leaf page
40 batches, when MMU sets it (e.g., on x86
41 theoretically worsen lock contention (m
42 disabled, the multi-gen LRU will suffer
43 degradation for workloads that contiguo
44 whose accessed bits can be otherwise cl
45 batches.
46 0x0004 Clearing the accessed bit in non-leaf p
47 well, when MMU sets it (e.g., on x86).
48 verified on x86 varieties other than In
49 disabled, the multi-gen LRU will suffer
50 performance degradation.
51 [yYnN] Apply to all the components above.
52 ====== =======================================
53
54 E.g.,
55 ::
56
57 echo y >/sys/kernel/mm/lru_gen/enabled
58 cat /sys/kernel/mm/lru_gen/enabled
59 0x0007
60 echo 5 >/sys/kernel/mm/lru_gen/enabled
61 cat /sys/kernel/mm/lru_gen/enabled
62 0x0005
63
64 Thrashing prevention
65 --------------------
66 Personal computers are more sensitive to thras
67 cause janks (lags when rendering UI) and negat
68 experience. The multi-gen LRU offers thrashing
69 majority of laptop and desktop users who do no
70
71 Users can write ``N`` to ``min_ttl_ms`` to pre
72 ``N`` milliseconds from getting evicted. The O
73 if this working set cannot be kept in memory.
74 option works as an adjustable pressure relief
75 terminates applications that are hopefully not
76
77 Based on the average human detectable lag (~10
78 eliminates intolerable janks due to thrashing.
79 ``N=3000`` make janks less noticeable at the r
80 kills.
81
82 The default value ``0`` means disabled.
83
84 Experimental features
85 =====================
86 ``/sys/kernel/debug/lru_gen`` accepts commands
87 following subsections. Multiple command lines
88 concatenation with delimiters ``,`` and ``;``.
89
90 ``/sys/kernel/debug/lru_gen_full`` provides ad
91 debugging. ``CONFIG_LRU_GEN_STATS=y`` keeps hi
92 evicted generations in this file.
93
94 Working set estimation
95 ----------------------
96 Working set estimation measures how much memor
97 in a given time interval, and it is usually do
98 the performance of the application. E.g., data
99 optimize job scheduling (bin packing) to impro
100 When a new job comes in, the job scheduler nee
101 each server it manages can allocate a certain
102 this new job before it can pick a candidate. T
103 scheduler needs to estimate the working sets o
104
105 When it is read, ``lru_gen`` returns a histogr
106 accessed over different time intervals for eac
107 ``MAX_NR_GENS`` decides the number of bins for
108 histograms are noncumulative.
109 ::
110
111 memcg memcg_id memcg_path
112 node node_id
113 min_gen_nr age_in_ms nr_anon_page
114 ...
115 max_gen_nr age_in_ms nr_anon_page
116
117 Each bin contains an estimated number of pages
118 within ``age_in_ms``. E.g., ``min_gen_nr`` con
119 and ``max_gen_nr`` contains the hottest pages,
120 the former is the largest and that of the latt
121
122 Users can write the following command to ``lru
123 generation ``max_gen_nr+1``:
124
125 ``+ memcg_id node_id max_gen_nr [can_swap
126
127 ``can_swap`` defaults to the swap setting and,
128 it forces the scan of anon pages when swap is
129 ``force_scan`` defaults to ``1`` and, if it is
130 employs heuristics to reduce the overhead, whi
131 the coverage as well.
132
133 A typical use case is that a job scheduler run
134 certain time interval to create new generation
135 servers it manages based on the sizes of their
136 this time interval.
137
138 Proactive reclaim
139 -----------------
140 Proactive reclaim induces page reclaim when th
141 pressure. It usually targets cold pages only.
142 comes in, the job scheduler wants to proactive
143 the server it selected, to improve the chance
144 this new job.
145
146 Users can write the following command to ``lru
147 generations less than or equal to ``min_gen_nr
148
149 ``- memcg_id node_id min_gen_nr [swappines
150
151 ``min_gen_nr`` should be less than ``max_gen_n
152 ``max_gen_nr`` and ``max_gen_nr-1`` are not fu
153 the active list) and therefore cannot be evict
154 overrides the default value in ``/proc/sys/vm/
155 ``nr_to_reclaim`` limits the number of pages t
156
157 A typical use case is that a job scheduler run
158 tries to land a new job on a server. If it fai
159 cold pages because of the overestimation, it r
160 server according to the ranking result obtaine
161 estimation step. This less forceful approach l
162 existing jobs.
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.