iostats.txt 7.1 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150
  1. I/O statistics fields
  2. ---------------
  3. Last modified Sep 30, 2003
  4. Since 2.4.20 (and some versions before, with patches), and 2.5.45,
  5. more extensive disk statistics have been introduced to help measure disk
  6. activity. Tools such as sar and iostat typically interpret these and do
  7. the work for you, but in case you are interested in creating your own
  8. tools, the fields are explained here.
  9. In 2.4 now, the information is found as additional fields in
  10. /proc/partitions. In 2.6, the same information is found in two
  11. places: one is in the file /proc/diskstats, and the other is within
  12. the sysfs file system, which must be mounted in order to obtain
  13. the information. Throughout this document we'll assume that sysfs
  14. is mounted on /sys, although of course it may be mounted anywhere.
  15. Both /proc/diskstats and sysfs use the same source for the information
  16. and so should not differ.
  17. Here are examples of these different formats:
  18. 2.4:
  19. 3 0 39082680 hda 446216 784926 9550688 4382310 424847 312726 5922052 19310380 0 3376340 23705160
  20. 3 1 9221278 hda1 35486 0 35496 38030 0 0 0 0 0 38030 38030
  21. 2.6 sysfs:
  22. 446216 784926 9550688 4382310 424847 312726 5922052 19310380 0 3376340 23705160
  23. 35486 38030 38030 38030
  24. 2.6 diskstats:
  25. 3 0 hda 446216 784926 9550688 4382310 424847 312726 5922052 19310380 0 3376340 23705160
  26. 3 1 hda1 35486 38030 38030 38030
  27. On 2.4 you might execute "grep 'hda ' /proc/partitions". On 2.6, you have
  28. a choice of "cat /sys/block/hda/stat" or "grep 'hda ' /proc/diskstats".
  29. The advantage of one over the other is that the sysfs choice works well
  30. if you are watching a known, small set of disks. /proc/diskstats may
  31. be a better choice if you are watching a large number of disks because
  32. you'll avoid the overhead of 50, 100, or 500 or more opens/closes with
  33. each snapshot of your disk statistics.
  34. In 2.4, the statistics fields are those after the device name. In
  35. the above example, the first field of statistics would be 446216.
  36. By contrast, in 2.6 if you look at /sys/block/hda/stat, you'll
  37. find just the eleven fields, beginning with 446216. If you look at
  38. /proc/diskstats, the eleven fields will be preceded by the major and
  39. minor device numbers, and device name. Each of these formats provide
  40. eleven fields of statistics, each meaning exactly the same things.
  41. All fields except field 9 are cumulative since boot. Field 9 should
  42. go to zero as I/Os complete; all others only increase. Yes, these are
  43. 32 bit unsigned numbers, and on a very busy or long-lived system they
  44. may wrap. Applications should be prepared to deal with that; unless
  45. your observations are measured in large numbers of minutes or hours,
  46. they should not wrap twice before you notice them.
  47. Each set of stats only applies to the indicated device; if you want
  48. system-wide stats you'll have to find all the devices and sum them all up.
  49. Field 1 -- # of reads issued
  50. This is the total number of reads completed successfully.
  51. Field 2 -- # of reads merged, field 6 -- # of writes merged
  52. Reads and writes which are adjacent to each other may be merged for
  53. efficiency. Thus two 4K reads may become one 8K read before it is
  54. ultimately handed to the disk, and so it will be counted (and queued)
  55. as only one I/O. This field lets you know how often this was done.
  56. Field 3 -- # of sectors read
  57. This is the total number of sectors read successfully.
  58. Field 4 -- # of milliseconds spent reading
  59. This is the total number of milliseconds spent by all reads (as
  60. measured from __make_request() to end_that_request_last()).
  61. Field 5 -- # of writes completed
  62. This is the total number of writes completed successfully.
  63. Field 7 -- # of sectors written
  64. This is the total number of sectors written successfully.
  65. Field 8 -- # of milliseconds spent writing
  66. This is the total number of milliseconds spent by all writes (as
  67. measured from __make_request() to end_that_request_last()).
  68. Field 9 -- # of I/Os currently in progress
  69. The only field that should go to zero. Incremented as requests are
  70. given to appropriate request_queue_t and decremented as they finish.
  71. Field 10 -- # of milliseconds spent doing I/Os
  72. This field is increases so long as field 9 is nonzero.
  73. Field 11 -- weighted # of milliseconds spent doing I/Os
  74. This field is incremented at each I/O start, I/O completion, I/O
  75. merge, or read of these stats by the number of I/Os in progress
  76. (field 9) times the number of milliseconds spent doing I/O since the
  77. last update of this field. This can provide an easy measure of both
  78. I/O completion time and the backlog that may be accumulating.
  79. To avoid introducing performance bottlenecks, no locks are held while
  80. modifying these counters. This implies that minor inaccuracies may be
  81. introduced when changes collide, so (for instance) adding up all the
  82. read I/Os issued per partition should equal those made to the disks ...
  83. but due to the lack of locking it may only be very close.
  84. In 2.6, there are counters for each cpu, which made the lack of locking
  85. almost a non-issue. When the statistics are read, the per-cpu counters
  86. are summed (possibly overflowing the unsigned 32-bit variable they are
  87. summed to) and the result given to the user. There is no convenient
  88. user interface for accessing the per-cpu counters themselves.
  89. Disks vs Partitions
  90. -------------------
  91. There were significant changes between 2.4 and 2.6 in the I/O subsystem.
  92. As a result, some statistic information disappeared. The translation from
  93. a disk address relative to a partition to the disk address relative to
  94. the host disk happens much earlier. All merges and timings now happen
  95. at the disk level rather than at both the disk and partition level as
  96. in 2.4. Consequently, you'll see a different statistics output on 2.6 for
  97. partitions from that for disks. There are only *four* fields available
  98. for partitions on 2.6 machines. This is reflected in the examples above.
  99. Field 1 -- # of reads issued
  100. This is the total number of reads issued to this partition.
  101. Field 2 -- # of sectors read
  102. This is the total number of sectors requested to be read from this
  103. partition.
  104. Field 3 -- # of writes issued
  105. This is the total number of writes issued to this partition.
  106. Field 4 -- # of sectors written
  107. This is the total number of sectors requested to be written to
  108. this partition.
  109. Note that since the address is translated to a disk-relative one, and no
  110. record of the partition-relative address is kept, the subsequent success
  111. or failure of the read cannot be attributed to the partition. In other
  112. words, the number of reads for partitions is counted slightly before time
  113. of queuing for partitions, and at completion for whole disks. This is
  114. a subtle distinction that is probably uninteresting for most cases.
  115. Additional notes
  116. ----------------
  117. In 2.6, sysfs is not mounted by default. If your distribution of
  118. Linux hasn't added it already, here's the line you'll want to add to
  119. your /etc/fstab:
  120. none /sys sysfs defaults 0 0
  121. In 2.6, all disk statistics were removed from /proc/stat. In 2.4, they
  122. appear in both /proc/partitions and /proc/stat, although the ones in
  123. /proc/stat take a very different format from those in /proc/partitions
  124. (see proc(5), if your system has it.)
  125. -- ricklind@us.ibm.com