Back to home page

LXR

 
 

    


0001 I/O statistics fields
0002 ---------------
0003 
0004 Since 2.4.20 (and some versions before, with patches), and 2.5.45,
0005 more extensive disk statistics have been introduced to help measure disk
0006 activity. Tools such as sar and iostat typically interpret these and do
0007 the work for you, but in case you are interested in creating your own
0008 tools, the fields are explained here.
0009 
0010 In 2.4 now, the information is found as additional fields in
0011 /proc/partitions.  In 2.6, the same information is found in two
0012 places: one is in the file /proc/diskstats, and the other is within
0013 the sysfs file system, which must be mounted in order to obtain
0014 the information. Throughout this document we'll assume that sysfs
0015 is mounted on /sys, although of course it may be mounted anywhere.
0016 Both /proc/diskstats and sysfs use the same source for the information
0017 and so should not differ.
0018 
0019 Here are examples of these different formats:
0020 
0021 2.4:
0022    3     0   39082680 hda 446216 784926 9550688 4382310 424847 312726 5922052 19310380 0 3376340 23705160
0023    3     1    9221278 hda1 35486 0 35496 38030 0 0 0 0 0 38030 38030
0024 
0025 
0026 2.6 sysfs:
0027    446216 784926 9550688 4382310 424847 312726 5922052 19310380 0 3376340 23705160
0028    35486    38030    38030    38030
0029 
0030 2.6 diskstats:
0031    3    0   hda 446216 784926 9550688 4382310 424847 312726 5922052 19310380 0 3376340 23705160
0032    3    1   hda1 35486 38030 38030 38030
0033 
0034 On 2.4 you might execute "grep 'hda ' /proc/partitions". On 2.6, you have
0035 a choice of "cat /sys/block/hda/stat" or "grep 'hda ' /proc/diskstats".
0036 The advantage of one over the other is that the sysfs choice works well
0037 if you are watching a known, small set of disks.  /proc/diskstats may
0038 be a better choice if you are watching a large number of disks because
0039 you'll avoid the overhead of 50, 100, or 500 or more opens/closes with
0040 each snapshot of your disk statistics.
0041 
0042 In 2.4, the statistics fields are those after the device name. In
0043 the above example, the first field of statistics would be 446216.
0044 By contrast, in 2.6 if you look at /sys/block/hda/stat, you'll
0045 find just the eleven fields, beginning with 446216.  If you look at
0046 /proc/diskstats, the eleven fields will be preceded by the major and
0047 minor device numbers, and device name.  Each of these formats provides
0048 eleven fields of statistics, each meaning exactly the same things.
0049 All fields except field 9 are cumulative since boot.  Field 9 should
0050 go to zero as I/Os complete; all others only increase (unless they
0051 overflow and wrap).  Yes, these are (32-bit or 64-bit) unsigned long
0052 (native word size) numbers, and on a very busy or long-lived system they
0053 may wrap. Applications should be prepared to deal with that; unless
0054 your observations are measured in large numbers of minutes or hours,
0055 they should not wrap twice before you notice them.
0056 
0057 Each set of stats only applies to the indicated device; if you want
0058 system-wide stats you'll have to find all the devices and sum them all up.
0059 
0060 Field  1 -- # of reads completed
0061     This is the total number of reads completed successfully.
0062 Field  2 -- # of reads merged, field 6 -- # of writes merged
0063     Reads and writes which are adjacent to each other may be merged for
0064     efficiency.  Thus two 4K reads may become one 8K read before it is
0065     ultimately handed to the disk, and so it will be counted (and queued)
0066     as only one I/O.  This field lets you know how often this was done.
0067 Field  3 -- # of sectors read
0068     This is the total number of sectors read successfully.
0069 Field  4 -- # of milliseconds spent reading
0070     This is the total number of milliseconds spent by all reads (as
0071     measured from __make_request() to end_that_request_last()).
0072 Field  5 -- # of writes completed
0073     This is the total number of writes completed successfully.
0074 Field  6 -- # of writes merged
0075     See the description of field 2.
0076 Field  7 -- # of sectors written
0077     This is the total number of sectors written successfully.
0078 Field  8 -- # of milliseconds spent writing
0079     This is the total number of milliseconds spent by all writes (as
0080     measured from __make_request() to end_that_request_last()).
0081 Field  9 -- # of I/Os currently in progress
0082     The only field that should go to zero. Incremented as requests are
0083     given to appropriate struct request_queue and decremented as they finish.
0084 Field 10 -- # of milliseconds spent doing I/Os
0085     This field increases so long as field 9 is nonzero.
0086 Field 11 -- weighted # of milliseconds spent doing I/Os
0087     This field is incremented at each I/O start, I/O completion, I/O
0088     merge, or read of these stats by the number of I/Os in progress
0089     (field 9) times the number of milliseconds spent doing I/O since the
0090     last update of this field.  This can provide an easy measure of both
0091     I/O completion time and the backlog that may be accumulating.
0092 
0093 
0094 To avoid introducing performance bottlenecks, no locks are held while
0095 modifying these counters.  This implies that minor inaccuracies may be
0096 introduced when changes collide, so (for instance) adding up all the
0097 read I/Os issued per partition should equal those made to the disks ...
0098 but due to the lack of locking it may only be very close.
0099 
0100 In 2.6, there are counters for each CPU, which make the lack of locking
0101 almost a non-issue.  When the statistics are read, the per-CPU counters
0102 are summed (possibly overflowing the unsigned long variable they are
0103 summed to) and the result given to the user.  There is no convenient
0104 user interface for accessing the per-CPU counters themselves.
0105 
0106 Disks vs Partitions
0107 -------------------
0108 
0109 There were significant changes between 2.4 and 2.6 in the I/O subsystem.
0110 As a result, some statistic information disappeared. The translation from
0111 a disk address relative to a partition to the disk address relative to
0112 the host disk happens much earlier.  All merges and timings now happen
0113 at the disk level rather than at both the disk and partition level as
0114 in 2.4.  Consequently, you'll see a different statistics output on 2.6 for
0115 partitions from that for disks.  There are only *four* fields available
0116 for partitions on 2.6 machines.  This is reflected in the examples above.
0117 
0118 Field  1 -- # of reads issued
0119     This is the total number of reads issued to this partition.
0120 Field  2 -- # of sectors read
0121     This is the total number of sectors requested to be read from this
0122     partition.
0123 Field  3 -- # of writes issued
0124     This is the total number of writes issued to this partition.
0125 Field  4 -- # of sectors written
0126     This is the total number of sectors requested to be written to
0127     this partition.
0128 
0129 Note that since the address is translated to a disk-relative one, and no
0130 record of the partition-relative address is kept, the subsequent success
0131 or failure of the read cannot be attributed to the partition.  In other
0132 words, the number of reads for partitions is counted slightly before time
0133 of queuing for partitions, and at completion for whole disks.  This is
0134 a subtle distinction that is probably uninteresting for most cases.
0135 
0136 More significant is the error induced by counting the numbers of
0137 reads/writes before merges for partitions and after for disks. Since a
0138 typical workload usually contains a lot of successive and adjacent requests,
0139 the number of reads/writes issued can be several times higher than the
0140 number of reads/writes completed.
0141 
0142 In 2.6.25, the full statistic set is again available for partitions and
0143 disk and partition statistics are consistent again. Since we still don't
0144 keep record of the partition-relative address, an operation is attributed to
0145 the partition which contains the first sector of the request after the
0146 eventual merges. As requests can be merged across partition, this could lead
0147 to some (probably insignificant) inaccuracy.
0148 
0149 Additional notes
0150 ----------------
0151 
0152 In 2.6, sysfs is not mounted by default.  If your distribution of
0153 Linux hasn't added it already, here's the line you'll want to add to
0154 your /etc/fstab:
0155 
0156 none /sys sysfs defaults 0 0
0157 
0158 
0159 In 2.6, all disk statistics were removed from /proc/stat.  In 2.4, they
0160 appear in both /proc/partitions and /proc/stat, although the ones in
0161 /proc/stat take a very different format from those in /proc/partitions
0162 (see proc(5), if your system has it.)
0163 
0164 -- ricklind@us.ibm.com