1I/O statistics fields
2---------------
3
4Since 2.4.20 (and some versions before, with patches), and 2.5.45,
5more extensive disk statistics have been introduced to help measure disk
6activity. Tools such as sar and iostat typically interpret these and do
7the work for you, but in case you are interested in creating your own
8tools, the fields are explained here.
9
10In 2.4 now, the information is found as additional fields in
11/proc/partitions.  In 2.6, the same information is found in two
12places: one is in the file /proc/diskstats, and the other is within
13the sysfs file system, which must be mounted in order to obtain
14the information. Throughout this document we'll assume that sysfs
15is mounted on /sys, although of course it may be mounted anywhere.
16Both /proc/diskstats and sysfs use the same source for the information
17and so should not differ.
18
19Here are examples of these different formats:
20
212.4:
22   3     0   39082680 hda 446216 784926 9550688 4382310 424847 312726 5922052 19310380 0 3376340 23705160
23   3     1    9221278 hda1 35486 0 35496 38030 0 0 0 0 0 38030 38030
24
25
262.6 sysfs:
27   446216 784926 9550688 4382310 424847 312726 5922052 19310380 0 3376340 23705160
28   35486    38030    38030    38030
29
302.6 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".
36The advantage of one over the other is that the sysfs choice works well
37if you are watching a known, small set of disks.  /proc/diskstats may
38be a better choice if you are watching a large number of disks because
39you'll avoid the overhead of 50, 100, or 500 or more opens/closes with
40each snapshot of your disk statistics.
41
42In 2.4, the statistics fields are those after the device name. In
43the above example, the first field of statistics would be 446216.
44By contrast, in 2.6 if you look at /sys/block/hda/stat, you'll
45find just the eleven fields, beginning with 446216.  If you look at
46/proc/diskstats, the eleven fields will be preceded by the major and
47minor device numbers, and device name.  Each of these formats provides
48eleven fields of statistics, each meaning exactly the same things.
49All fields except field 9 are cumulative since boot.  Field 9 should
50go to zero as I/Os complete; all others only increase (unless they
51overflow and wrap).  Yes, these are (32-bit or 64-bit) unsigned long
52(native word size) numbers, and on a very busy or long-lived system they
53may wrap. Applications should be prepared to deal with that; unless
54your observations are measured in large numbers of minutes or hours,
55they should not wrap twice before you notice them.
56
57Each set of stats only applies to the indicated device; if you want
58system-wide stats you'll have to find all the devices and sum them all up.
59
60Field  1 -- # of reads completed
61    This is the total number of reads completed successfully.
62Field  2 -- # of reads merged, field 6 -- # of writes merged
63    Reads and writes which are adjacent to each other may be merged for
64    efficiency.  Thus two 4K reads may become one 8K read before it is
65    ultimately handed to the disk, and so it will be counted (and queued)
66    as only one I/O.  This field lets you know how often this was done.
67Field  3 -- # of sectors read
68    This is the total number of sectors read successfully.
69Field  4 -- # of milliseconds spent reading
70    This is the total number of milliseconds spent by all reads (as
71    measured from __make_request() to end_that_request_last()).
72Field  5 -- # of writes completed
73    This is the total number of writes completed successfully.
74Field  6 -- # of writes merged
75    See the description of field 2.
76Field  7 -- # of sectors written
77    This is the total number of sectors written successfully.
78Field  8 -- # of milliseconds spent writing
79    This is the total number of milliseconds spent by all writes (as
80    measured from __make_request() to end_that_request_last()).
81Field  9 -- # of I/Os currently in progress
82    The only field that should go to zero. Incremented as requests are
83    given to appropriate struct request_queue and decremented as they finish.
84Field 10 -- # of milliseconds spent doing I/Os
85    This field increases so long as field 9 is nonzero.
86Field 11 -- weighted # of milliseconds spent doing I/Os
87    This field is incremented at each I/O start, I/O completion, I/O
88    merge, or read of these stats by the number of I/Os in progress
89    (field 9) times the number of milliseconds spent doing I/O since the
90    last update of this field.  This can provide an easy measure of both
91    I/O completion time and the backlog that may be accumulating.
92
93
94To avoid introducing performance bottlenecks, no locks are held while
95modifying these counters.  This implies that minor inaccuracies may be
96introduced when changes collide, so (for instance) adding up all the
97read I/Os issued per partition should equal those made to the disks ...
98but due to the lack of locking it may only be very close.
99
100In 2.6, there are counters for each CPU, which make the lack of locking
101almost a non-issue.  When the statistics are read, the per-CPU counters
102are summed (possibly overflowing the unsigned long variable they are
103summed to) and the result given to the user.  There is no convenient
104user interface for accessing the per-CPU counters themselves.
105
106Disks vs Partitions
107-------------------
108
109There were significant changes between 2.4 and 2.6 in the I/O subsystem.
110As a result, some statistic information disappeared. The translation from
111a disk address relative to a partition to the disk address relative to
112the host disk happens much earlier.  All merges and timings now happen
113at the disk level rather than at both the disk and partition level as
114in 2.4.  Consequently, you'll see a different statistics output on 2.6 for
115partitions from that for disks.  There are only *four* fields available
116for partitions on 2.6 machines.  This is reflected in the examples above.
117
118Field  1 -- # of reads issued
119    This is the total number of reads issued to this partition.
120Field  2 -- # of sectors read
121    This is the total number of sectors requested to be read from this
122    partition.
123Field  3 -- # of writes issued
124    This is the total number of writes issued to this partition.
125Field  4 -- # of sectors written
126    This is the total number of sectors requested to be written to
127    this partition.
128
129Note that since the address is translated to a disk-relative one, and no
130record of the partition-relative address is kept, the subsequent success
131or failure of the read cannot be attributed to the partition.  In other
132words, the number of reads for partitions is counted slightly before time
133of queuing for partitions, and at completion for whole disks.  This is
134a subtle distinction that is probably uninteresting for most cases.
135
136More significant is the error induced by counting the numbers of
137reads/writes before merges for partitions and after for disks. Since a
138typical workload usually contains a lot of successive and adjacent requests,
139the number of reads/writes issued can be several times higher than the
140number of reads/writes completed.
141
142In 2.6.25, the full statistic set is again available for partitions and
143disk and partition statistics are consistent again. Since we still don't
144keep record of the partition-relative address, an operation is attributed to
145the partition which contains the first sector of the request after the
146eventual merges. As requests can be merged across partition, this could lead
147to some (probably insignificant) inaccuracy.
148
149Additional notes
150----------------
151
152In 2.6, sysfs is not mounted by default.  If your distribution of
153Linux hasn't added it already, here's the line you'll want to add to
154your /etc/fstab:
155
156none /sys sysfs defaults 0 0
157
158
159In 2.6, all disk statistics were removed from /proc/stat.  In 2.4, they
160appear in both /proc/partitions and /proc/stat, although the ones in
161/proc/stat take a very different format from those in /proc/partitions
162(see proc(5), if your system has it.)
163
164-- ricklind@us.ibm.com
165