1 =====================
2 I/O statistics fields
3 =====================
4
5 Since 2.4.20 (and some versions before, with p
6 more extensive disk statistics have been intro
7 activity. Tools such as ``sar`` and ``iostat``
8 the work for you, but in case you are interest
9 tools, the fields are explained here.
10
11 In 2.4 now, the information is found as additi
12 ``/proc/partitions``. In 2.6 and upper, the s
13 places: one is in the file ``/proc/diskstats``
14 the sysfs file system, which must be mounted i
15 the information. Throughout this document we'l
16 is mounted on ``/sys``, although of course it
17 Both ``/proc/diskstats`` and sysfs use the sam
18 and so should not differ.
19
20 Here are examples of these different formats::
21
22 2.4:
23 3 0 39082680 hda 446216 784926 955
24 3 1 9221278 hda1 35486 0 35496 38
25
26 2.6+ sysfs:
27 446216 784926 9550688 4382310 424847 312
28 35486 38030 38030 38030
29
30 2.6+ diskstats:
31 3 0 hda 446216 784926 9550688 43823
32 3 1 hda1 35486 38030 38030 38030
33
34 4.18+ diskstats:
35 3 0 hda 446216 784926 9550688 43823
36
37 On 2.4 you might execute ``grep 'hda ' /proc/p
38 a choice of ``cat /sys/block/hda/stat`` or ``g
39
40 The advantage of one over the other is that th
41 if you are watching a known, small set of disk
42 be a better choice if you are watching a large
43 you'll avoid the overhead of 50, 100, or 500 o
44 each snapshot of your disk statistics.
45
46 In 2.4, the statistics fields are those after
47 the above example, the first field of statisti
48 By contrast, in 2.6+ if you look at ``/sys/blo
49 find just the 15 fields, beginning with 446216
50 ``/proc/diskstats``, the 15 fields will be pre
51 minor device numbers, and device name. Each o
52 15 fields of statistics, each meaning exactly
53 All fields except field 9 are cumulative since
54 go to zero as I/Os complete; all others only i
55 overflow and wrap). Wrapping might eventually
56 or long-lived system; so applications should b
57 it. Regarding wrapping, the types of the field
58 int (32 bit) or unsigned long (32-bit or 64-bi
59 machine) as noted per-field below. Unless your
60 spread in time, these fields should not wrap t
61
62 Each set of stats only applies to the indicate
63 system-wide stats you'll have to find all the
64
65 Field 1 -- # of reads completed (unsigned lon
66 This is the total number of reads complete
67
68 Field 2 -- # of reads merged, field 6 -- # of
69 Reads and writes which are adjacent to eac
70 efficiency. Thus two 4K reads may become
71 ultimately handed to the disk, and so it w
72 as only one I/O. This field lets you know
73
74 Field 3 -- # of sectors read (unsigned long)
75 This is the total number of sectors read s
76
77 Field 4 -- # of milliseconds spent reading (u
78 This is the total number of milliseconds s
79 measured from blk_mq_alloc_request() to __
80
81 Field 5 -- # of writes completed (unsigned lo
82 This is the total number of writes complet
83
84 Field 6 -- # of writes merged (unsigned long
85 See the description of field 2.
86
87 Field 7 -- # of sectors written (unsigned lon
88 This is the total number of sectors writte
89
90 Field 8 -- # of milliseconds spent writing (u
91 This is the total number of milliseconds s
92 measured from blk_mq_alloc_request() to __
93
94 Field 9 -- # of I/Os currently in progress (u
95 The only field that should go to zero. Inc
96 given to appropriate struct request_queue
97
98 Field 10 -- # of milliseconds spent doing I/Os
99 This field increases so long as field 9 is
100
101 Since 5.0 this field counts jiffies when a
102 started or completed. If request runs more
103 I/O time might be not accounted in case of
104
105 Field 11 -- weighted # of milliseconds spent d
106 This field is incremented at each I/O star
107 merge, or read of these stats by the numbe
108 (field 9) times the number of milliseconds
109 last update of this field. This can provi
110 I/O completion time and the backlog that m
111
112 Field 12 -- # of discards completed (unsigned
113 This is the total number of discards compl
114
115 Field 13 -- # of discards merged (unsigned lon
116 See the description of field 2
117
118 Field 14 -- # of sectors discarded (unsigned l
119 This is the total number of sectors discar
120
121 Field 15 -- # of milliseconds spent discarding
122 This is the total number of milliseconds s
123 measured from blk_mq_alloc_request() to __
124
125 Field 16 -- # of flush requests completed
126 This is the total number of flush requests
127
128 Block layer combines flush requests and ex
129 This counts flush requests executed by dis
130
131 Field 17 -- # of milliseconds spent flushing
132 This is the total number of milliseconds s
133
134 To avoid introducing performance bottlenecks,
135 modifying these counters. This implies that m
136 introduced when changes collide, so (for insta
137 read I/Os issued per partition should equal th
138 but due to the lack of locking it may only be
139
140 In 2.6+, there are counters for each CPU, whic
141 almost a non-issue. When the statistics are r
142 are summed (possibly overflowing the unsigned
143 summed to) and the result given to the user.
144 user interface for accessing the per-CPU count
145
146 Since 4.19 request times are measured with nan
147 truncated to milliseconds before showing in th
148
149 Disks vs Partitions
150 -------------------
151
152 There were significant changes between 2.4 and
153 As a result, some statistic information disapp
154 a disk address relative to a partition to the
155 the host disk happens much earlier. All merge
156 at the disk level rather than at both the disk
157 in 2.4. Consequently, you'll see a different
158 partitions from that for disks. There are onl
159 for partitions on 2.6+ machines. This is refl
160
161 Field 1 -- # of reads issued
162 This is the total number of reads issued t
163
164 Field 2 -- # of sectors read
165 This is the total number of sectors reques
166 partition.
167
168 Field 3 -- # of writes issued
169 This is the total number of writes issued
170
171 Field 4 -- # of sectors written
172 This is the total number of sectors reques
173 this partition.
174
175 Note that since the address is translated to a
176 record of the partition-relative address is ke
177 or failure of the read cannot be attributed to
178 words, the number of reads for partitions is c
179 of queuing for partitions, and at completion f
180 a subtle distinction that is probably unintere
181
182 More significant is the error induced by count
183 reads/writes before merges for partitions and
184 typical workload usually contains a lot of suc
185 the number of reads/writes issued can be sever
186 number of reads/writes completed.
187
188 In 2.6.25, the full statistic set is again ava
189 disk and partition statistics are consistent a
190 keep record of the partition-relative address,
191 the partition which contains the first sector
192 eventual merges. As requests can be merged acr
193 to some (probably insignificant) inaccuracy.
194
195 Additional notes
196 ----------------
197
198 In 2.6+, sysfs is not mounted by default. If
199 Linux hasn't added it already, here's the line
200 your ``/etc/fstab``::
201
202 none /sys sysfs defaults 0 0
203
204
205 In 2.6+, all disk statistics were removed from
206 appear in both ``/proc/partitions`` and ``/pro
207 ``/proc/stat`` take a very different format fr
208 (see proc(5), if your system has it.)
209
210 -- ricklind@us.ibm.com
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.