1 ================= 1 =================
2 Scheduler debugfs 2 Scheduler debugfs
3 ================= 3 =================
4 4
5 Booting a kernel with CONFIG_SCHED_DEBUG=y wil 5 Booting a kernel with CONFIG_SCHED_DEBUG=y will give access to
6 scheduler specific debug files under /sys/kern 6 scheduler specific debug files under /sys/kernel/debug/sched. Some of
7 those files are described below. 7 those files are described below.
8 8
9 numa_balancing 9 numa_balancing
10 ============== 10 ==============
11 11
12 `numa_balancing` directory is used to hold fil 12 `numa_balancing` directory is used to hold files to control NUMA
13 balancing feature. If the system overhead fro 13 balancing feature. If the system overhead from the feature is too
14 high then the rate the kernel samples for NUMA 14 high then the rate the kernel samples for NUMA hinting faults may be
15 controlled by the `scan_period_min_ms, scan_de 15 controlled by the `scan_period_min_ms, scan_delay_ms,
16 scan_period_max_ms, scan_size_mb` files. 16 scan_period_max_ms, scan_size_mb` files.
17 17
18 18
19 scan_period_min_ms, scan_delay_ms, scan_period 19 scan_period_min_ms, scan_delay_ms, scan_period_max_ms, scan_size_mb
20 ---------------------------------------------- 20 -------------------------------------------------------------------
21 21
22 Automatic NUMA balancing scans tasks address s 22 Automatic NUMA balancing scans tasks address space and unmaps pages to
23 detect if pages are properly placed or if the 23 detect if pages are properly placed or if the data should be migrated to a
24 memory node local to where the task is running 24 memory node local to where the task is running. Every "scan delay" the task
25 scans the next "scan size" number of pages in 25 scans the next "scan size" number of pages in its address space. When the
26 end of the address space is reached the scanne 26 end of the address space is reached the scanner restarts from the beginning.
27 27
28 In combination, the "scan delay" and "scan siz 28 In combination, the "scan delay" and "scan size" determine the scan rate.
29 When "scan delay" decreases, the scan rate inc 29 When "scan delay" decreases, the scan rate increases. The scan delay and
30 hence the scan rate of every task is adaptive 30 hence the scan rate of every task is adaptive and depends on historical
31 behaviour. If pages are properly placed then t 31 behaviour. If pages are properly placed then the scan delay increases,
32 otherwise the scan delay decreases. The "scan 32 otherwise the scan delay decreases. The "scan size" is not adaptive but
33 the higher the "scan size", the higher the sca 33 the higher the "scan size", the higher the scan rate.
34 34
35 Higher scan rates incur higher system overhead 35 Higher scan rates incur higher system overhead as page faults must be
36 trapped and potentially data must be migrated. 36 trapped and potentially data must be migrated. However, the higher the scan
37 rate, the more quickly a tasks memory is migra 37 rate, the more quickly a tasks memory is migrated to a local node if the
38 workload pattern changes and minimises perform 38 workload pattern changes and minimises performance impact due to remote
39 memory accesses. These files control the thres 39 memory accesses. These files control the thresholds for scan delays and
40 the number of pages scanned. 40 the number of pages scanned.
41 41
42 ``scan_period_min_ms`` is the minimum time in 42 ``scan_period_min_ms`` is the minimum time in milliseconds to scan a
43 tasks virtual memory. It effectively controls 43 tasks virtual memory. It effectively controls the maximum scanning
44 rate for each task. 44 rate for each task.
45 45
46 ``scan_delay_ms`` is the starting "scan delay" 46 ``scan_delay_ms`` is the starting "scan delay" used for a task when it
47 initially forks. 47 initially forks.
48 48
49 ``scan_period_max_ms`` is the maximum time in 49 ``scan_period_max_ms`` is the maximum time in milliseconds to scan a
50 tasks virtual memory. It effectively controls 50 tasks virtual memory. It effectively controls the minimum scanning
51 rate for each task. 51 rate for each task.
52 52
53 ``scan_size_mb`` is how many megabytes worth o 53 ``scan_size_mb`` is how many megabytes worth of pages are scanned for
54 a given scan. 54 a given scan.
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.