linux/Documentation/iostats.txt
<<
>>
Prefs
   1=====================
   2I/O statistics fields
   3=====================
   4
   5Since 2.4.20 (and some versions before, with patches), and 2.5.45,
   6more extensive disk statistics have been introduced to help measure disk
   7activity. Tools such as ``sar`` and ``iostat`` typically interpret these and do
   8the work for you, but in case you are interested in creating your own
   9tools, the fields are explained here.
  10
  11In 2.4 now, the information is found as additional fields in
  12``/proc/partitions``.  In 2.6 and upper, the same information is found in two
  13places: one is in the file ``/proc/diskstats``, and the other is within
  14the sysfs file system, which must be mounted in order to obtain
  15the information. Throughout this document we'll assume that sysfs
  16is mounted on ``/sys``, although of course it may be mounted anywhere.
  17Both ``/proc/diskstats`` and sysfs use the same source for the information
  18and so should not differ.
  19
  20Here are examples of these different formats::
  21
  22   2.4:
  23      3     0   39082680 hda 446216 784926 9550688 4382310 424847 312726 5922052 19310380 0 3376340 23705160
  24      3     1    9221278 hda1 35486 0 35496 38030 0 0 0 0 0 38030 38030
  25
  26   2.6+ sysfs:
  27      446216 784926 9550688 4382310 424847 312726 5922052 19310380 0 3376340 23705160
  28      35486    38030    38030    38030
  29
  30   2.6+ diskstats:
  31      3    0   hda 446216 784926 9550688 4382310 424847 312726 5922052 19310380 0 3376340 23705160
  32      3    1   hda1 35486 38030 38030 38030
  33
  34On 2.4 you might execute ``grep 'hda ' /proc/partitions``. On 2.6+, you have
  35a choice of ``cat /sys/block/hda/stat`` or ``grep 'hda ' /proc/diskstats``.
  36
  37The advantage of one over the other is that the sysfs choice works well
  38if you are watching a known, small set of disks.  ``/proc/diskstats`` may
  39be a better choice if you are watching a large number of disks because
  40you'll avoid the overhead of 50, 100, or 500 or more opens/closes with
  41each snapshot of your disk statistics.
  42
  43In 2.4, the statistics fields are those after the device name. In
  44the above example, the first field of statistics would be 446216.
  45By contrast, in 2.6+ if you look at ``/sys/block/hda/stat``, you'll
  46find just the eleven fields, beginning with 446216.  If you look at
  47``/proc/diskstats``, the eleven fields will be preceded by the major and
  48minor device numbers, and device name.  Each of these formats provides
  49eleven fields of statistics, each meaning exactly the same things.
  50All fields except field 9 are cumulative since boot.  Field 9 should
  51go to zero as I/Os complete; all others only increase (unless they
  52overflow and wrap).  Yes, these are (32-bit or 64-bit) unsigned long
  53(native word size) numbers, and on a very busy or long-lived system they
  54may wrap. Applications should be prepared to deal with that; unless
  55your observations are measured in large numbers of minutes or hours,
  56they should not wrap twice before you notice them.
  57
  58Each set of stats only applies to the indicated device; if you want
  59system-wide stats you'll have to find all the devices and sum them all up.
  60
  61Field  1 -- # of reads completed
  62    This is the total number of reads completed successfully.
  63
  64Field  2 -- # of reads merged, field 6 -- # of writes merged
  65    Reads and writes which are adjacent to each other may be merged for
  66    efficiency.  Thus two 4K reads may become one 8K read before it is
  67    ultimately handed to the disk, and so it will be counted (and queued)
  68    as only one I/O.  This field lets you know how often this was done.
  69
  70Field  3 -- # of sectors read
  71    This is the total number of sectors read successfully.
  72
  73Field  4 -- # of milliseconds spent reading
  74    This is the total number of milliseconds spent by all reads (as
  75    measured from __make_request() to end_that_request_last()).
  76
  77Field  5 -- # of writes completed
  78    This is the total number of writes completed successfully.
  79
  80Field  6 -- # of writes merged
  81    See the description of field 2.
  82
  83Field  7 -- # of sectors written
  84    This is the total number of sectors written successfully.
  85
  86Field  8 -- # of milliseconds spent writing
  87    This is the total number of milliseconds spent by all writes (as
  88    measured from __make_request() to end_that_request_last()).
  89
  90Field  9 -- # of I/Os currently in progress
  91    The only field that should go to zero. Incremented as requests are
  92    given to appropriate struct request_queue and decremented as they finish.
  93
  94Field 10 -- # of milliseconds spent doing I/Os
  95    This field increases so long as field 9 is nonzero.
  96
  97Field 11 -- weighted # of milliseconds spent doing I/Os
  98    This field is incremented at each I/O start, I/O completion, I/O
  99    merge, or read of these stats by the number of I/Os in progress
 100    (field 9) times the number of milliseconds spent doing I/O since the
 101    last update of this field.  This can provide an easy measure of both
 102    I/O completion time and the backlog that may be accumulating.
 103
 104
 105To avoid introducing performance bottlenecks, no locks are held while
 106modifying these counters.  This implies that minor inaccuracies may be
 107introduced when changes collide, so (for instance) adding up all the
 108read I/Os issued per partition should equal those made to the disks ...
 109but due to the lack of locking it may only be very close.
 110
 111In 2.6+, there are counters for each CPU, which make the lack of locking
 112almost a non-issue.  When the statistics are read, the per-CPU counters
 113are summed (possibly overflowing the unsigned long variable they are
 114summed to) and the result given to the user.  There is no convenient
 115user interface for accessing the per-CPU counters themselves.
 116
 117Disks vs Partitions
 118-------------------
 119
 120There were significant changes between 2.4 and 2.6+ in the I/O subsystem.
 121As a result, some statistic information disappeared. The translation from
 122a disk address relative to a partition to the disk address relative to
 123the host disk happens much earlier.  All merges and timings now happen
 124at the disk level rather than at both the disk and partition level as
 125in 2.4.  Consequently, you'll see a different statistics output on 2.6+ for
 126partitions from that for disks.  There are only *four* fields available
 127for partitions on 2.6+ machines.  This is reflected in the examples above.
 128
 129Field  1 -- # of reads issued
 130    This is the total number of reads issued to this partition.
 131
 132Field  2 -- # of sectors read
 133    This is the total number of sectors requested to be read from this
 134    partition.
 135
 136Field  3 -- # of writes issued
 137    This is the total number of writes issued to this partition.
 138
 139Field  4 -- # of sectors written
 140    This is the total number of sectors requested to be written to
 141    this partition.
 142
 143Note that since the address is translated to a disk-relative one, and no
 144record of the partition-relative address is kept, the subsequent success
 145or failure of the read cannot be attributed to the partition.  In other
 146words, the number of reads for partitions is counted slightly before time
 147of queuing for partitions, and at completion for whole disks.  This is
 148a subtle distinction that is probably uninteresting for most cases.
 149
 150More significant is the error induced by counting the numbers of
 151reads/writes before merges for partitions and after for disks. Since a
 152typical workload usually contains a lot of successive and adjacent requests,
 153the number of reads/writes issued can be several times higher than the
 154number of reads/writes completed.
 155
 156In 2.6.25, the full statistic set is again available for partitions and
 157disk and partition statistics are consistent again. Since we still don't
 158keep record of the partition-relative address, an operation is attributed to
 159the partition which contains the first sector of the request after the
 160eventual merges. As requests can be merged across partition, this could lead
 161to some (probably insignificant) inaccuracy.
 162
 163Additional notes
 164----------------
 165
 166In 2.6+, sysfs is not mounted by default.  If your distribution of
 167Linux hasn't added it already, here's the line you'll want to add to
 168your ``/etc/fstab``::
 169
 170        none /sys sysfs defaults 0 0
 171
 172
 173In 2.6+, all disk statistics were removed from ``/proc/stat``.  In 2.4, they
 174appear in both ``/proc/partitions`` and ``/proc/stat``, although the ones in
 175``/proc/stat`` take a very different format from those in ``/proc/partitions``
 176(see proc(5), if your system has it.)
 177
 178-- ricklind@us.ibm.com
 179