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
34   4.18+ diskstats:
35      3    0   hda 446216 784926 9550688 4382310 424847 312726 5922052 19310380 0 3376340 23705160 0 0 0 0
36
37On 2.4 you might execute ``grep 'hda ' /proc/partitions``. On 2.6+, you have
38a choice of ``cat /sys/block/hda/stat`` or ``grep 'hda ' /proc/diskstats``.
39
40The advantage of one over the other is that the sysfs choice works well
41if you are watching a known, small set of disks.  ``/proc/diskstats`` may
42be a better choice if you are watching a large number of disks because
43you'll avoid the overhead of 50, 100, or 500 or more opens/closes with
44each snapshot of your disk statistics.
45
46In 2.4, the statistics fields are those after the device name. In
47the above example, the first field of statistics would be 446216.
48By contrast, in 2.6+ if you look at ``/sys/block/hda/stat``, you'll
49find just the 15 fields, beginning with 446216.  If you look at
50``/proc/diskstats``, the 15 fields will be preceded by the major and
51minor device numbers, and device name.  Each of these formats provides
5215 fields of statistics, each meaning exactly the same things.
53All fields except field 9 are cumulative since boot.  Field 9 should
54go to zero as I/Os complete; all others only increase (unless they
55overflow and wrap). Wrapping might eventually occur on a very busy
56or long-lived system; so applications should be prepared to deal with
57it. Regarding wrapping, the types of the fields are either unsigned
58int (32 bit) or unsigned long (32-bit or 64-bit, depending on your
59machine) as noted per-field below. Unless your observations are very
60spread in time, these fields should not wrap twice before you notice it.
61
62Each set of stats only applies to the indicated device; if you want
63system-wide stats you'll have to find all the devices and sum them all up.
64
65Field  1 -- # of reads completed (unsigned long)
66    This is the total number of reads completed successfully.
67
68Field  2 -- # of reads merged, field 6 -- # of writes merged (unsigned long)
69    Reads and writes which are adjacent to each other may be merged for
70    efficiency.  Thus two 4K reads may become one 8K read before it is
71    ultimately handed to the disk, and so it will be counted (and queued)
72    as only one I/O.  This field lets you know how often this was done.
73
74Field  3 -- # of sectors read (unsigned long)
75    This is the total number of sectors read successfully.
76
77Field  4 -- # of milliseconds spent reading (unsigned int)
78    This is the total number of milliseconds spent by all reads (as
79    measured from __make_request() to end_that_request_last()).
80
81Field  5 -- # of writes completed (unsigned long)
82    This is the total number of writes completed successfully.
83
84Field  6 -- # of writes merged  (unsigned long)
85    See the description of field 2.
86
87Field  7 -- # of sectors written (unsigned long)
88    This is the total number of sectors written successfully.
89
90Field  8 -- # of milliseconds spent writing (unsigned int)
91    This is the total number of milliseconds spent by all writes (as
92    measured from __make_request() to end_that_request_last()).
93
94Field  9 -- # of I/Os currently in progress (unsigned int)
95    The only field that should go to zero. Incremented as requests are
96    given to appropriate struct request_queue and decremented as they finish.
97
98Field 10 -- # of milliseconds spent doing I/Os (unsigned int)
99    This field increases so long as field 9 is nonzero.
100
101    Since 5.0 this field counts jiffies when at least one request was
102    started or completed. If request runs more than 2 jiffies then some
103    I/O time will not be accounted unless there are other requests.
104
105Field 11 -- weighted # of milliseconds spent doing I/Os (unsigned int)
106    This field is incremented at each I/O start, I/O completion, I/O
107    merge, or read of these stats by the number of I/Os in progress
108    (field 9) times the number of milliseconds spent doing I/O since the
109    last update of this field.  This can provide an easy measure of both
110    I/O completion time and the backlog that may be accumulating.
111
112Field 12 -- # of discards completed (unsigned long)
113    This is the total number of discards completed successfully.
114
115Field 13 -- # of discards merged (unsigned long)
116    See the description of field 2
117
118Field 14 -- # of sectors discarded (unsigned long)
119    This is the total number of sectors discarded successfully.
120
121Field 15 -- # of milliseconds spent discarding (unsigned int)
122    This is the total number of milliseconds spent by all discards (as
123    measured from __make_request() to end_that_request_last()).
124
125Field 16 -- # of flush requests completed
126    This is the total number of flush requests completed successfully.
127
128    Block layer combines flush requests and executes at most one at a time.
129    This counts flush requests executed by disk. Not tracked for partitions.
130
131Field 17 -- # of milliseconds spent flushing
132    This is the total number of milliseconds spent by all flush requests.
133
134To avoid introducing performance bottlenecks, no locks are held while
135modifying these counters.  This implies that minor inaccuracies may be
136introduced when changes collide, so (for instance) adding up all the
137read I/Os issued per partition should equal those made to the disks ...
138but due to the lack of locking it may only be very close.
139
140In 2.6+, there are counters for each CPU, which make the lack of locking
141almost a non-issue.  When the statistics are read, the per-CPU counters
142are summed (possibly overflowing the unsigned long variable they are
143summed to) and the result given to the user.  There is no convenient
144user interface for accessing the per-CPU counters themselves.
145
146Disks vs Partitions
147-------------------
148
149There were significant changes between 2.4 and 2.6+ in the I/O subsystem.
150As a result, some statistic information disappeared. The translation from
151a disk address relative to a partition to the disk address relative to
152the host disk happens much earlier.  All merges and timings now happen
153at the disk level rather than at both the disk and partition level as
154in 2.4.  Consequently, you'll see a different statistics output on 2.6+ for
155partitions from that for disks.  There are only *four* fields available
156for partitions on 2.6+ machines.  This is reflected in the examples above.
157
158Field  1 -- # of reads issued
159    This is the total number of reads issued to this partition.
160
161Field  2 -- # of sectors read
162    This is the total number of sectors requested to be read from this
163    partition.
164
165Field  3 -- # of writes issued
166    This is the total number of writes issued to this partition.
167
168Field  4 -- # of sectors written
169    This is the total number of sectors requested to be written to
170    this partition.
171
172Note that since the address is translated to a disk-relative one, and no
173record of the partition-relative address is kept, the subsequent success
174or failure of the read cannot be attributed to the partition.  In other
175words, the number of reads for partitions is counted slightly before time
176of queuing for partitions, and at completion for whole disks.  This is
177a subtle distinction that is probably uninteresting for most cases.
178
179More significant is the error induced by counting the numbers of
180reads/writes before merges for partitions and after for disks. Since a
181typical workload usually contains a lot of successive and adjacent requests,
182the number of reads/writes issued can be several times higher than the
183number of reads/writes completed.
184
185In 2.6.25, the full statistic set is again available for partitions and
186disk and partition statistics are consistent again. Since we still don't
187keep record of the partition-relative address, an operation is attributed to
188the partition which contains the first sector of the request after the
189eventual merges. As requests can be merged across partition, this could lead
190to some (probably insignificant) inaccuracy.
191
192Additional notes
193----------------
194
195In 2.6+, sysfs is not mounted by default.  If your distribution of
196Linux hasn't added it already, here's the line you'll want to add to
197your ``/etc/fstab``::
198
199	none /sys sysfs defaults 0 0
200
201
202In 2.6+, all disk statistics were removed from ``/proc/stat``.  In 2.4, they
203appear in both ``/proc/partitions`` and ``/proc/stat``, although the ones in
204``/proc/stat`` take a very different format from those in ``/proc/partitions``
205(see proc(5), if your system has it.)
206
207-- ricklind@us.ibm.com
208