Linux cache命中率查看

一、需求与背景

LINUX主机上的内存很为user、buffer、cache这个大的块。经常在使用free -m查看时，会发现cache占去了大头，从理论上我们也一直都知道cache里存放的是系统上访问的热数据（不经常访问的数据会被标记后，更换为新的数据），这样会加快主机的访问速度。而cache内部这些数据的命中率如何查看呢？今天在跟相关人员分析现网一个问题时，有同事提到的一个问题。

二、cache命中率计算方法

网上找了下Brendan Gregg 在 perf-tools 里集成了一个工具cachestat，可以查看cache命中率，该工具是用shell 写的，读了下源码，发现其是通过读取/sys/kernel/debug/tracing/trace_stat/ 目录下的所有function调用，查看其中的四项内容：

 

1. mark_page_accessed() for measuring cache accesses
2. mark_buffer_dirty() for measuring cache writes
3. add_to_page_cache_lru() for measuring page additions
4. account_page_dirtied() for measuring page dirties

而四者之间的值是按如下公式计算的：

 

1. 总cache page数（total） = mark_page_accessed(标记为访问的) - mark_buffer_dirty（标记为脏的）
2. misses = add_to_page_cache_lru(新增的) - account_page_dirtied （统计为脏状态的）
3. total - misses = hit
4. hit/total = ratio (命中率)
5.  

这里给一个读取人trace_stat下的文件的内容，如下：

在该工具同时还读取了/proc/meminfo，不过读取该文件并不参与命中率计算，读取这个文件的内容只是用于算出目前使用的cache大小，并在界面输出。

三、cachestat代码及执行

cachestat代码如下：

 

1. #!/bin/bash
2. # 28-Dec-2014 Brendan Gregg Created this.
3. ### default variables
4. tracing=/sys/kernel/debug/tracing
5. interval=1; opt_timestamp=0; opt_debug=0
6. trap 'quit=1' INT QUIT TERM PIPE HUP # sends execution to end tracing section
7. function usage {
8. cat <<-END >&2
9. USAGE: cachestat [-Dht] [interval]
10. -D # print debug counters
11. -h # this usage message
12. -t # include timestamp
13. interval # output interval in secs (default 1)
14. eg,
15. cachestat # show stats every second
16. cachestat 5 # show stats every 5 seconds
17. See the man page and example file for more info.
18. END
19. exit
20. }
21. function warn {
22. if ! eval "$@"; then
23. echo >&2 "WARNING: command failed \"$@\""
24. fi
25. }
26. function die {
27. echo >&2 "$@"
28. exit 1
29. }
30. ### process options
31. while getopts Dht opt
32. do
33. case $opt in
34. D) opt_debug=1 ;;
35. t) opt_timestamp=1 ;;
36. h|?) usage ;;
37. esac
38. done
39. shift $(( $OPTIND - 1 ))
40. ### option logic
41. if (( $# )); then
42. interval=$1
43. fi
44. echo "Counting cache functions... Output every $interval seconds."
45. ### check permissions
46. cd $tracing || die "ERROR: accessing tracing. Root user? Kernel has FTRACE?
47. debugfs mounted? (mount -t debugfs debugfs /sys/kernel/debug)"
48. ### enable tracing
49. sysctl -q kernel.ftrace_enabled=1 # doesn't set exit status
50. printf "mark_page_accessed\nmark_buffer_dirty\nadd_to_page_cache_lru\naccount_page_dirtied\n" > set_ftrace_filter || \
51. die "ERROR: tracing these four kernel functions: mark_page_accessed,"\
52. "mark_buffer_dirty, add_to_page_cache_lru and account_page_dirtied (unknown kernel version?). Exiting."
53. warn "echo nop > current_tracer"
54. if ! echo 1 > function_profile_enabled; then
55. echo > set_ftrace_filter
56. die "ERROR: enabling function profiling. Have CONFIG_FUNCTION_PROFILER? Exiting."
57. fi
58. (( opt_timestamp )) && printf "%-8s " TIME
59. printf "%8s %8s %8s %8s %12s %10s" HITS MISSES DIRTIES RATIO "BUFFERS_MB" "CACHE_MB"
60. (( opt_debug )) && printf " DEBUG"
61. echo
62. ### summarize
63. quit=0; secs=0
64. while (( !quit && (!opt_duration || secs < duration) )); do
65. (( secs += interval ))
66. echo 0 > function_profile_enabled
67. echo 1 > function_profile_enabled
68. sleep $interval
69. (( opt_timestamp )) && printf "%(%H:%M:%S)T " -1
70. # cat both meminfo and trace stats, and let awk pick them apart
71. cat /proc/meminfo trace_stat/function* | awk -v debug=$opt_debug '
72. # match meminfo stats:
73. $1 == "Buffers:" && $3 == "kB" { buffers_mb = $2 / 1024 }
74. $1 == "Cached:" && $3 == "kB" { cached_mb = $2 / 1024 }
75. # identify and save trace counts:
76. $2 ~ /[0-9]/ && $3 != "kB" { a[$1] += $2 }
77. END {
78. mpa = a["mark_page_accessed"]
79. mbd = a["mark_buffer_dirty"]
80. apcl = a["add_to_page_cache_lru"]
81. apd = a["account_page_dirtied"]
82. total = mpa - mbd
83. misses = apcl - apd
84. if (misses < 0)
85. misses = 0
86. hits = total - misses
87. ratio = 100 * hits / total
88. printf "%8d %8d %8d %7.1f%% %12.0f %10.0f", hits, misses, mbd,
89. ratio, buffers_mb, cached_mb
90. if (debug)
91. printf " (%d %d %d %d)", mpa, mbd, apcl, apd
92. printf "\n"
93. }'
94. done
95. ### end tracing
96. echo 2>/dev/null
97. echo "Ending tracing..." 2>/dev/null
98. warn "echo 0 > function_profile_enabled"
99. warn "echo > set_ftrace_filter"

脚本执行后，结果如下：

参考页面：

http://www.brendangregg.com/blog/2014-12-31/linux-page-cache-hit-ratio.html （Brendan Gregg大神的博客）

 

转载至http://www.361way.com/linux-cachestat/5894.html