grio_bandwidth(1M) grio_bandwidth(1M)
grio_bandwidth - determines guaranteed-rate I/O device bandwidth
grio_bandwidth -d device -s iosize [ -ruwV ] [ -t sampletime ]
grio_bandwidth -c controller -s iosize [ -ruwV ] [ -t sampletime ]
grio_bandwidth -R -s iosize [ -fruwV ] [ -t sampletime ]
grio_bandwidth is used to determine the number of I/O operations of a
given size that can be guaranteed to be performed by a given disk device,
RAID subsystem or SCSI controller in one second. It creates processes
which repeatedly issue I/O requests of the given size, and device seeks
of random lengths, to the specified devices. The length of time to
complete each request is recorded and this information is used to
determine the average number of I/O operations that were performed by the
device per second over the sample period. This number is degraded by 10%
to provide an approximation of the increase in I/O response time caused
by a heavily loaded system. In addition, grio_bandwidth keeps track of
the longest time to complete an I/O operation time over the sample period
and calculates the number of such "worst case" operations that can be
performed each second. The system administrator can use grio_bandwidth
to obtain the bandwidth information needed to create new entries in the
Guaranteed Rate I/O device bandwidth information file /etc/grio_disks.
The -d option is used to specify a single disk device. The disk name
could be specified in the /dev/[r]dsk alias name format, or in the format
reported back by -P, -C and -F options of the grio(1M) command. The -c
option is used to specify a single SCSI controller name. The name has to
be specified in the format reported back by -P, -C and -F options of the
grio(1M) command. This is also the same format used in
/etc/ioconfig.conf. The -R option will cause the utility to scan for all
RAID devices attached to the system. The -s option is used to specify
the size of the I/O operation in bytes. It must be specified, there is no
default value. The -t option is used to specify the length of the sample
period in seconds. By default the sample period has a length of 10
minutes. If the -u option is specified, the /etc/grio_disks file will be
automatically updated with the device bandwidth information. The -f
option can only be used when both the -R and-u options are used. It
indicates that when creating the entries in the /etc//grio_disk file, the
RAID controller bandwidth should be split evening among all RAID luns
attached to the controller. By default, the RAID controller bandwidth
will be allocated to individual RAID luns on a first-come, first-served
basis. The user can use the -r and -w options to specify if device read
and/or write operations should be performed. By default only read
operations will be performed on the device. Normally, both -r and -w
should be specified so that the utility uses a mix of read and write
operations when determining the bandwidth characteristics. If it is
known that on the device most Guaranteed Rate I/O rate guarantees will be
used for performing either read or write operations, then the correct
option may be specified to obtain a more accurate bandwidth
characterization for the expected workload. The -V option causes the
Page 1
grio_bandwidth(1M) grio_bandwidth(1M)
utility to produce more verbose output. The utility will use these
values to determine the bandwidth performance characteristics of the
device. In the case of a single disk device, the performance
characteristics of the device will be computed. In the case of a RAID
device, the performance characteristics of the RAID controllers, as well
as the RAID luns, will be computed.
There are a number of factors to consider when using grio_bandwidth to
determine the GRIO bandwidth characteristics of a RAID device. First,
the grio_bandwidth utility invokes the raidcli(5) utility to determine
the RAID devices attached to the system. In order to run the raidcli
program successfully, the system must be a registered RAID agent.
Second, it is important to re-run the utility each time the RAID device
is reconfigured. The RAID device is a complex mechanism and
configuration changes will have performance impact. Morover, in order to
insure that the GRIO bandwidth rate guarantees can be met even when a
RAID disk has failed, the bandwidth number must be computed when the
device is in degraded mode. When running in degraded mode (ie. one of
the disk devices in each of the RAID luns is removed/missing), the RAID
controllers on the RAID must compute the data for the missing disk. This
will increase I/O request response time.
/etc/grio_disks
ggd(1M), grio(1M), grio_disks(4), grio(5), raidcli(5), ioconfig(1M).
PPPPaaaaggggeeee 2222 [ Back ]
|
