ZFS uses an ARC (adaptive replacement cache) which is not accounted for in the traditional Linux "cache" memory usage. How can I determine the current size as well as size boundaries of the ZFS ARC, and how do these relate to the amount of free or cache memory reported for example by free?
Asked
Active
Viewed 5.3k times
32
user
- 29,449
- 11
- 99
- 144
-
1For FreeBSD: [sysutils/zfs-stats](https://www.freshports.org/sysutils/zfs-stats/) – Graham Perrin Sep 28 '21 at 06:34
2 Answers
38
The ZFS code reports various statistics through procfs. To determine the size of the ARC, look at /proc/spl/kstat/zfs/arcstats (assuming procfs mounted on /proc, as customary), specifically the values for c, c_max and size. (See also this post on the Oracle Community forum. Alternative Internet Archive copy in case the Oracle site becomes unavailable.)
cis the target size of the ARC in bytesc_maxis the maximum size of the ARC in bytessizeis the current size of the ARC in bytes
The maximum size of the ARC can be adjusted either by passing a zfs_arc_max=N parameter to the zfs module (through modprobe), where N is the maximum ARC size in bytes, or on the fly by writing the new maximum size in bytes to /sys/module/zfs/parameters/zfs_arc_max.
Because of how ZFS on Linux is implemented, the ARC memory behaves like cache memory (for example, it is evicted if the system comes under memory pressure), but is aggregated by the kernel as ordinary memory allocations. This can lead to confusion as the system appears to have far less free memory than would be expected given the current system workload, but is normal.
To get the ARC size in megabytes, you can use something like awk '/^size/ { print $1 " " $3 / 1048576 }' < /proc/spl/kstat/zfs/arcstats. (1,048,576 is the number of bytes to the megabyte.)
For example, my system (which uses ZFS almost exclusively) might report
$ free -m
total used free shared buffers cached
Mem: 32194 17948 14245 0 158 982
-/+ buffers/cache: 16808 15386
Swap: 49152 0 49152
$ awk '/^size/ { print $1 " " $3 / 1048576 }' < /proc/spl/kstat/zfs/arcstats
size 8138.73
$
which means that the actual memory usage by currently resident processes is approximately 8,669 MB (16,808 MB reduced by 8,139 MB).
user
- 29,449
- 11
- 99
- 144
26
To build on Michael Kjörling's answer, you can also use arc_summary.py.
Here you can see how the ARC is using half of my desktop's memory:
root@host:~# free -g
total used free shared buffers cached
Mem: 62 56 6 1 1 5
-/+ buffers/cache: 49 13
Swap: 7 0 7
root@host:~# arc_summary.py
------------------------------------------------------------------------
ZFS Subsystem Report Fri Feb 24 19:44:20 2017
ARC Summary: (HEALTHY)
Memory Throttle Count: 0
ARC Misc:
Deleted: 1.33m
Mutex Misses: 99
Evict Skips: 99
ARC Size: 98.13% 30.80 GiB
Target Size: (Adaptive) 100.00% 31.39 GiB
Min Size (Hard Limit): 0.10% 32.00 MiB
Max Size (High Water): 1004:1 31.39 GiB
ARC Size Breakdown:
Recently Used Cache Size: 84.25% 26.45 GiB
Frequently Used Cache Size: 15.75% 4.95 GiB
ARC Hash Breakdown:
Elements Max: 1.11m
Elements Current: 53.48% 592.56k
Collisions: 763.42k
Chain Max: 4
Chains: 19.62k
ARC Total accesses: 36.34m
Cache Hit Ratio: 87.02% 31.62m
Cache Miss Ratio: 12.98% 4.72m
Actual Hit Ratio: 84.78% 30.81m
Data Demand Efficiency: 93.49% 24.22m
Data Prefetch Efficiency: 2.57% 819.12k
CACHE HITS BY CACHE LIST:
Anonymously Used: 2.27% 716.60k
Most Recently Used: 17.26% 5.46m
Most Frequently Used: 80.17% 25.35m
Most Recently Used Ghost: 0.19% 60.25k
Most Frequently Used Ghost: 0.11% 35.37k
CACHE HITS BY DATA TYPE:
Demand Data: 71.60% 22.64m
Prefetch Data: 0.07% 21.04k
Demand Metadata: 25.82% 8.16m
Prefetch Metadata: 2.51% 794.39k
CACHE MISSES BY DATA TYPE:
Demand Data: 33.44% 1.58m
Prefetch Data: 16.92% 798.09k
Demand Metadata: 48.90% 2.31m
Prefetch Metadata: 0.75% 35.27k
DMU Prefetch Efficiency: 173.06m
Hit Ratio: 86.14% 149.07m
Miss Ratio: 13.86% 23.99m
ZFS Tunable:
metaslab_debug_load 0
zfs_arc_min_prefetch_lifespan 0
zfetch_max_streams 8
zfs_nopwrite_enabled 1
zfetch_min_sec_reap 2
zfs_dbgmsg_enable 0
zfs_dirty_data_max_max_percent 25
zfs_arc_p_aggressive_disable 1
spa_load_verify_data 1
zfs_zevent_cols 80
zfs_dirty_data_max_percent 10
zfs_sync_pass_dont_compress 5
l2arc_write_max 8388608
zfs_vdev_scrub_max_active 2
zfs_vdev_sync_write_min_active 10
zvol_prefetch_bytes 131072
metaslab_aliquot 524288
zfs_no_scrub_prefetch 0
zfs_arc_shrink_shift 0
zfetch_block_cap 256
zfs_txg_history 0
zfs_delay_scale 500000
zfs_vdev_async_write_active_min_dirty_percent 30
metaslab_debug_unload 0
zfs_read_history 0
zvol_max_discard_blocks 16384
zfs_recover 0
l2arc_headroom 2
zfs_deadman_synctime_ms 1000000
zfs_scan_idle 50
zfs_free_min_time_ms 1000
zfs_dirty_data_max 6741298790
zfs_vdev_async_read_min_active 1
zfs_mg_noalloc_threshold 0
zfs_dedup_prefetch 0
zfs_vdev_max_active 1000
l2arc_write_boost 8388608
zfs_resilver_min_time_ms 3000
zfs_vdev_async_write_max_active 10
zil_slog_limit 1048576
zfs_prefetch_disable 0
zfs_resilver_delay 2
metaslab_lba_weighting_enabled 1
zfs_mg_fragmentation_threshold 85
l2arc_feed_again 1
zfs_zevent_console 0
zfs_immediate_write_sz 32768
zfs_dbgmsg_maxsize 4194304
zfs_free_leak_on_eio 0
zfs_deadman_enabled 1
metaslab_bias_enabled 1
zfs_arc_p_dampener_disable 1
zfs_object_mutex_size 64
zfs_metaslab_fragmentation_threshold 70
zfs_no_scrub_io 0
metaslabs_per_vdev 200
zfs_dbuf_state_index 0
zfs_vdev_sync_read_min_active 10
metaslab_fragmentation_factor_enabled 1
zvol_inhibit_dev 0
zfs_vdev_async_write_active_max_dirty_percent 60
zfs_vdev_cache_size 0
zfs_vdev_mirror_switch_us 10000
zfs_dirty_data_sync 67108864
spa_config_path /etc/zfs/zpool.cache
zfs_dirty_data_max_max 16853246976
zfs_arc_lotsfree_percent 10
zfs_zevent_len_max 128
zfs_scan_min_time_ms 1000
zfs_arc_sys_free 0
zfs_arc_meta_strategy 1
zfs_vdev_cache_bshift 16
zfs_arc_meta_adjust_restarts 4096
zfs_max_recordsize 1048576
zfs_vdev_scrub_min_active 1
zfs_vdev_read_gap_limit 32768
zfs_arc_meta_limit 0
zfs_vdev_sync_write_max_active 10
l2arc_norw 0
zfs_arc_meta_prune 10000
metaslab_preload_enabled 1
l2arc_nocompress 0
zvol_major 230
zfs_vdev_aggregation_limit 131072
zfs_flags 0
spa_asize_inflation 24
zfs_admin_snapshot 0
l2arc_feed_secs 1
zio_taskq_batch_pct 75
zfs_sync_pass_deferred_free 2
zfs_disable_dup_eviction 0
zfs_arc_grow_retry 0
zfs_read_history_hits 0
zfs_vdev_async_write_min_active 1
zfs_vdev_async_read_max_active 3
zfs_scrub_delay 4
zfs_delay_min_dirty_percent 60
zfs_free_max_blocks 100000
zfs_vdev_cache_max 16384
zio_delay_max 30000
zfs_top_maxinflight 32
ignore_hole_birth 1
spa_slop_shift 5
zfs_vdev_write_gap_limit 4096
spa_load_verify_metadata 1
spa_load_verify_maxinflight 10000
l2arc_noprefetch 1
zfs_vdev_scheduler noop
zfs_expire_snapshot 300
zfs_sync_pass_rewrite 2
zil_replay_disable 0
zfs_nocacheflush 0
zfs_arc_max 0
zfs_arc_min 0
zfs_read_chunk_size 1048576
zfs_txg_timeout 5
zfs_pd_bytes_max 52428800
l2arc_headroom_boost 200
zfs_send_corrupt_data 0
l2arc_feed_min_ms 200
zfs_arc_meta_min 0
zfs_arc_average_blocksize 8192
zfetch_array_rd_sz 1048576
zfs_autoimport_disable 1
zfs_arc_p_min_shift 0
zio_requeue_io_start_cut_in_line 1
zfs_vdev_sync_read_max_active 10
zfs_mdcomp_disable 0
zfs_arc_num_sublists_per_state 8
11181
- 517
- 4
- 5
-
I wonder what the significance of the ratio reported for Max Size (High Water) means. – CMCDragonkai Aug 11 '17 at 11:52
-
10