dkstat(1M) dkstat(1M)
NAME
dkstat - report disk I/O statistics
SYNOPSIS
dkstat [-abcdgpqrstvx] [dkname ...]
dkstat -z [dkname ...]
dkstat interval [count] [dkname ...]
dkstat -L interval [count] [dkname ...]
DESCRIPTION
dkstat displays various I/O statistics collected per disk since boot-
time, including scaled profiles of block size, disk queue length, seek
distance, cylinder use, partition use, disk response time, sort queue
time, response time, load average and total usage characteristics.
Current disk usage can be monitored by specifying an interval in
seconds. A continuous display of average usage over each interval is
produced.
You must select an option or specify an interval. dkstat will not work
correctly without an option or interval.
Statistics can be cleared with the -z flag. If you use the -z option
with other options, the other options will be ignored; therefore,
always use the -z option on a separate line.
OPTIONS
-a all statistical profiles.
-b a profile of the number of blocks that were requested in each
I/O.
-c profile of the cylinders on which I/O occurred.
-d profile of the response time in milliseconds from the sending of
the I/O to the device and the receipt of the completion inter-
rupt.
-g read, write and byte counts. The read and write counts are the
total number of requests of each type. The byte count is the
total number of bytes transferred in both read and write
requests.
-p profile of the disk partitions on which I/O occurred.
-q profile of the disk request queue length at the time each request
is added to the queue.
-r profile of the response time in milliseconds from the issue of
the I/O to the receipt of the completion interrupt. This is the
sum of the values displayed by the -d and -t flags.
Page 1 Reliant UNIX 5.44 Printed 11/98
dkstat(1M) dkstat(1M)
-s profile of the absolute value of the distance sought in
cylinders.
-t profile of the response time in milliseconds from the time the
I/O is added to the queue to the time the I/O is sent to the
device.
-v forces the display of all data. If the -v flag is not used,
dkstat will truncate trailing zeros and print ellipses at the end
of non-zero data.
-x prints a summary of the options.
-z clears all statistics.
-L display load average over interval.
The optional trailing interval causes dkstat to report the average
disk transfers per second on each disk over that interval repeated an
optional count times. If count is omitted, the output is repeated
continuously. Drives may be specified with an optional trailing
dkname. Names are given in either the 13 character device name form or
an abbreviated form which includes bus number, bus type, and con-
troller and device numbers. For example, ios0/sdisk001 or s001. The
shell metacharacter syntax "?", "*", and "[" may also be used. For
example, ios0/sdisk* ?disk[01]??, or even *[01]. These names should
be escaped on shell level where necessary.
If no drives are specified, then the five disks with the highest usage
since boot will be displayed. If the -L load average flag is speci-
fied, then the four disks with highest usage during the interval are
displayed.
Profile information is collected from several arrays within the kernel
address space. Each array has two logical areas with selectable scal-
ing of data. These arrays are described by maximum values of both seg-
ments and the scale factor within both segments.
Averages are generated which are based on scaling data with the begin-
ning value of the sample cell or the end value of the sample cell. The
true average can be extrapolated between these values. Because the
data is collected in cells, the exact average cannot be determined.
The load average is a number indicating the overall effect of a disk
on system I/O performance. While the details of I/O performance may be
obtained by carefully examining the profile information, the load
average is meant to give an overview of performance of each device in
a single number. The disk load average is calculated as an exponen-
tial reduction of the current load. The load is calculated for an
interval using the following formula:
load = devicebusytime * ( L1 + L2 )
Page 2 Reliant UNIX 5.44 Printed 11/98
dkstat(1M) dkstat(1M)
where:
L1 = newrequests/requestscompleted
L2 = averagequeuedepth
SEE ALSO
sar(1), mpstat(1M), vmstat(1M).
Page 3 Reliant UNIX 5.44 Printed 11/98