1I/O statistics fields
   4Last modified Sep 30, 2003
   6Since 2.4.20 (and some versions before, with patches), and 2.5.45,
   7more extensive disk statistics have been introduced to help measure disk
   8activity. Tools such as sar and iostat typically interpret these and do
   9the work for you, but in case you are interested in creating your own
  10tools, the fields are explained here.
  12In 2.4 now, the information is found as additional fields in
  13/proc/partitions.  In 2.6, the same information is found in two
  14places: one is in the file /proc/diskstats, and the other is within
  15the sysfs file system, which must be mounted in order to obtain
  16the information. Throughout this document we'll assume that sysfs
  17is mounted on /sys, although of course it may be mounted anywhere.
  18Both /proc/diskstats and sysfs use the same source for the information
  19and so should not differ.
  21Here are examples of these different formats:
  24   3     0   39082680 hda 446216 784926 9550688 4382310 424847 312726 5922052 19310380 0 3376340 23705160
  25   3     1    9221278 hda1 35486 0 35496 38030 0 0 0 0 0 38030 38030
  282.6 sysfs:
  29   446216 784926 9550688 4382310 424847 312726 5922052 19310380 0 3376340 23705160
  30   35486    38030    38030    38030
  322.6 diskstats:
  33   3    0   hda 446216 784926 9550688 4382310 424847 312726 5922052 19310380 0 3376340 23705160
  34   3    1   hda1 35486 38030 38030 38030
  36On 2.4 you might execute "grep 'hda ' /proc/partitions". On 2.6, you have
  37a choice of "cat /sys/block/hda/stat" or "grep 'hda ' /proc/diskstats".
  38The advantage of one over the other is that the sysfs choice works well
  39if you are watching a known, small set of disks.  /proc/diskstats may
  40be a better choice if you are watching a large number of disks because
  41you'll avoid the overhead of 50, 100, or 500 or more opens/closes with
  42each snapshot of your disk statistics.
  44In 2.4, the statistics fields are those after the device name. In
  45the above example, the first field of statistics would be 446216.
  46By contrast, in 2.6 if you look at /sys/block/hda/stat, you'll
  47find just the eleven fields, beginning with 446216.  If you look at
  48/proc/diskstats, the eleven fields will be preceded by the major and
  49minor device numbers, and device name.  Each of these formats provide
  50eleven fields of statistics, each meaning exactly the same things.
  51All fields except field 9 are cumulative since boot.  Field 9 should
  52go to zero as I/Os complete; all others only increase.  Yes, these are
  5332 bit unsigned 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.
  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.
  61Field  1 -- # of reads completed
  62    This is the total number of reads completed successfully.
  63Field  2 -- # of reads merged, field 6 -- # of writes merged
  64    Reads and writes which are adjacent to each other may be merged for
  65    efficiency.  Thus two 4K reads may become one 8K read before it is
  66    ultimately handed to the disk, and so it will be counted (and queued)
  67    as only one I/O.  This field lets you know how often this was done.
  68Field  3 -- # of sectors read
  69    This is the total number of sectors read successfully.
  70Field  4 -- # of milliseconds spent reading
  71    This is the total number of milliseconds spent by all reads (as
  72    measured from __make_request() to end_that_request_last()).
  73Field  5 -- # of writes completed
  74    This is the total number of writes completed successfully.
  75Field  7 -- # of sectors written
  76    This is the total number of sectors written successfully.
  77Field  8 -- # of milliseconds spent writing
  78    This is the total number of milliseconds spent by all writes (as
  79    measured from __make_request() to end_that_request_last()).
  80Field  9 -- # of I/Os currently in progress
  81    The only field that should go to zero. Incremented as requests are
  82    given to appropriate struct request_queue and decremented as they finish.
  83Field 10 -- # of milliseconds spent doing I/Os
  84    This field is increases so long as field 9 is nonzero.
  85Field 11 -- weighted # of milliseconds spent doing I/Os
  86    This field is incremented at each I/O start, I/O completion, I/O
  87    merge, or read of these stats by the number of I/Os in progress
  88    (field 9) times the number of milliseconds spent doing I/O since the
  89    last update of this field.  This can provide an easy measure of both
  90    I/O completion time and the backlog that may be accumulating.
  93To avoid introducing performance bottlenecks, no locks are held while
  94modifying these counters.  This implies that minor inaccuracies may be
  95introduced when changes collide, so (for instance) adding up all the
  96read I/Os issued per partition should equal those made to the disks ...
  97but due to the lack of locking it may only be very close.
  99In 2.6, there are counters for each cpu, which made the lack of locking
 100almost a non-issue.  When the statistics are read, the per-cpu counters
 101are summed (possibly overflowing the unsigned 32-bit variable they are
 102summed to) and the result given to the user.  There is no convenient
 103user interface for accessing the per-cpu counters themselves.
 105Disks vs Partitions
 108There were significant changes between 2.4 and 2.6 in the I/O subsystem.
 109As a result, some statistic information disappeared. The translation from
 110a disk address relative to a partition to the disk address relative to
 111the host disk happens much earlier.  All merges and timings now happen
 112at the disk level rather than at both the disk and partition level as
 113in 2.4.  Consequently, you'll see a different statistics output on 2.6 for
 114partitions from that for disks.  There are only *four* fields available
 115for partitions on 2.6 machines.  This is reflected in the examples above.
 117Field  1 -- # of reads issued
 118    This is the total number of reads issued to this partition.
 119Field  2 -- # of sectors read
 120    This is the total number of sectors requested to be read from this
 121    partition.
 122Field  3 -- # of writes issued
 123    This is the total number of writes issued to this partition.
 124Field  4 -- # of sectors written
 125    This is the total number of sectors requested to be written to
 126    this partition.
 128Note that since the address is translated to a disk-relative one, and no
 129record of the partition-relative address is kept, the subsequent success
 130or failure of the read cannot be attributed to the partition.  In other
 131words, the number of reads for partitions is counted slightly before time
 132of queuing for partitions, and at completion for whole disks.  This is
 133a subtle distinction that is probably uninteresting for most cases.
 135More significant is the error induced by counting the numbers of
 136reads/writes before merges for partitions and after for disks. Since a
 137typical workload usually contains a lot of successive and adjacent requests,
 138the number of reads/writes issued can be several times higher than the
 139number of reads/writes completed.
 141In 2.6.25, the full statistic set is again available for partitions and
 142disk and partition statistics are consistent again. Since we still don't
 143keep record of the partition-relative address, an operation is attributed to
 144the partition which contains the first sector of the request after the
 145eventual merges. As requests can be merged across partition, this could lead
 146to some (probably insignificant) inaccuracy.
 148Additional notes
 151In 2.6, sysfs is not mounted by default.  If your distribution of
 152Linux hasn't added it already, here's the line you'll want to add to
 153your /etc/fstab:
 155none /sys sysfs defaults 0 0
 158In 2.6, all disk statistics were removed from /proc/stat.  In 2.4, they
 159appear in both /proc/partitions and /proc/stat, although the ones in
 160/proc/stat take a very different format from those in /proc/partitions
 161(see proc(5), if your system has it.)