Configuring a balanced storage subsystem
This task provides step-by-step
instructions for configuring a balanced storage subsystem.
The attachment of a given storage subsystem to a SAN Volume Controller requires that some
specific settings be applied to the device, some limitations are also listed
for each storage type. There are 2 major steps in this process:
- Setting the characteristics of the SAN Volume Controller to storage connection(s)
- Mapping logical unit(s) to these connections such that the SAN Volume Controller can access
them.
The virtualization features of the IBM Total Storage SAN Volume Controller enable you to
choose how your storage is divided up and presented to hosts. While virtualization
provides you with a great deal of flexibility, it also offers the potential
for setting up a storage subsystem which can be overloaded. A storage subsystem
is overloaded if the quantity of I/O transactions that are being issued by
the host systems exceeds the capability of the storage to process those transactions.
If a storage subsystem is overloaded, then, at best, it causes delays in the
host systems and, at worst, it causes I/O transactions to be timed out in
the host which leads to errors being logged by the hosts and I/Os being failed
back to applications.
As an extreme example of an overloaded storage subsystem it would be possible
to use an IBM Total Storage SAN Volume Controller to virtualize a single RAID array and to
divide this storage among sixty-four host systems. Clearly, if all host systems
attempt to access this storage at the same time the single RAID array will
be overloaded. The following guidelines are provided to help you configure
balanced storage subsystems.
Steps
Perform the following steps to configure balanced storage
subsystems:
- Calculate the I/O rate for an array. For each RAID array in the storage
subsystem use the following table to calculate the approximate number of I/O
operations per second that can be processed by the RAID array. Note that the
actual number of I/O operations per second that can be processed will vary
depending on the location and length of each I/O, whether the I/O is a read
or a write operation and on the specifications of the component disks of the
RAID array.
Table 37. Calculate the I/O rate
| Type of RAID Array |
Number of component disks in the RAID
Array |
Approximate I/O rate |
| RAID-1 (mirrored) arrays |
2 |
300 |
| RAID-3, RAID-4, RAID-5 (striped + parity) arrays |
N + 1 parity |
150 * N |
| RAID-10, RAID 0+1, RAID 1+0 (striped + mirrored) arrays |
N |
150 * N |
For example, a RAID-5 array with eight component disks has an approximate
I/O rate of 150 * 7 = 1050.
- Calculate the I/O rate for a managed disk. If there is a one-to-one relationship
between back-end arrays and managed disks (this is the recommended configuration)
then the I/O rate for a managed disk is the same as the I/O rate of the corresponding
array. If an array is divided up into multiple managed disks then the I/O
rate per managed disk is the I/O rate of the array divided by the number of
managed disks that are using that array.
- Calculate the I/O rate for a managed disk group. The I/O rate for a managed
disk group is simply the sum of the I/O rates of the managed disk within that
group.
For example, a managed disk group contains eight managed disks each
of which corresponds to a RAID-1 array. From the table above the I/O rate
for each managed disks can be calculated as 300. The I/O rate for the managed
disk group is therefore 300 * 8 = 2400.
-
Calculate the impact of FlashCopy relationships. If you
are using the FlashCopy feature provided by the IBM Total Storage SAN Volume Controller then
you need to consider how much additional I/O will be generated by using this
feature as this will reduce the rate at which I/O from host systems that can
be processed. When a FlashCopy relationship is copying data any write I/Os
from host systems to areas of the source or target virtual disk that have
not yet been copied will cause extra I/Os to be generated by the IBM Total
Storage SAN Volume Controller to copy the data before the write I/O is actually performed.
The effect of using FlashCopy depends on the type of I/O workload being generated
by an application:
Table 38. Calculate the impact of FlashCopy relationships
| Type of application |
Impact to I/O rate |
Additional Weighting for FlashCopy |
| Application is doing no I/O |
Insignificant impact |
0 |
| Application is only reading data |
Insignificant impact |
0 |
| Application is only issuing random writes |
Up to 50 times as much I/O |
49 |
| Application is issuing random reads and writes |
Up to 15 times as much I/O |
14 |
| Application is issuing sequential reads or writes |
Up to 2 times as much I/O |
1 |
For each virtual disk that is either the source or target of an
active FlashCopy relationship consider the type of application that will be
using that virtual disk and record the additional weighting for that virtual
disk.
For example, a FlashCopy relationship is being used to provide
point in time backups. During the FlashCopy process, a host application generates
an I/O workload of random reads and writes to the source virtual disk. A second
host application reads the target virtual disk and writes the data to tape
to create a backup. The additional weighting for the source virtual disk is
14. The additional weighting for the destination virtual disk is 0.
- Calculate the I/O rate for virtual disks in a managed disk group. Calculate
the number of virtual disks in the managed disk group. Add the additional
weighting for each virtual disk that is either the source or a target of an
active FlashCopy relationship. Divide the I/O rate of the managed disk group
by this number to give an I/O rate per VDisk.
Example 1: A managed disk
group has an I/O rate of 2400 and contains 20 virtual disks. There are no
FlashCopy relationships. The I/O rate per virtual disk is 2400 / 20 = 120.
Example 2: A managed disk group has an I/O rate of 5000 and contains
20 virtual disks. There are two active FlashCopy relationships which have
source virtual disks in this managed disk group. Both source virtual disks
are being accessed by applications issuing random reads and write and hence
the additional weighting for each of these virtual disks is 14. The I/O rate
per virtual disk is 5000 / ( 20 + 14 + 14 ) = 104.
- Determine whether the storage subsystem is overloaded. The figure determined
at step 4 provides some indication of how many I/O
operations per second can be processed by each virtual disk in the managed
disk group. If you know how many I/O operations per second your
host applications generate you can compare these figures to determine if the
system is overloaded. If you do not know how many I/O operations per second
your host applications generate then you can either measure this (for example
by using the I/O statistics facilities provided by the IBM Total Storage SAN Volume Controller to
measure the I/O rate of your virtual disks) or use the following table as
a guideline:
Table 39. Determine if the storage subsystem is overloaded
| Type of Application |
I/O rate per virtual disk |
| Applications that generate a high I/O workload |
200 |
| Applications that generate a medium I/O workload |
80 |
| Applications that generate a low I/O workload |
10 |
- Interpret the result. If the I/O rate generated by the application exceeds
the I/O rate you calculated per virtual disk this indicates that you could
be overloading your storage subsystem and you should monitor the system carefully
to see if the back-end storage is actually limiting the overall performance
of your system. It is also possible that the calculation above is too simplistic
to model your use of storage, for example the calculation assumes that your
applications generate the same I/O workload to all virtual disks.
One method
you can use to monitor the performance of your storage subsystem is to use
the I/O statistics facilities provided by the IBM Total Storage SAN Volume Controller to measure
the I/O rate of your managed disks. Alternatively you could use the performance
and I/O statistics facilities provided by your back-end controllers.
If you find your storage subsystem is over loaded there are several actions
that can be take to resolve the problem:
- Adding more back-end storage to the system will allow you to increase
the quantity of I/O that can be processed by your storage subsystem. The virtualization
and data migration facilities provided by the IBM Total Storage SAN Volume Controller can
be used to redistribute the I/O work load of virtual disks across a greater
number of managed disks without having to take the storage offline.
- Stop any unessential FlashCopy relationships as this will reduce the amount
of I/O operations submitted to the back-end storage. If you are making many
FlashCopy's in parallel then consider starting less FlashCopy relationships
in parallel.
- The I/O workload generated by a host can often be limited by adjusting
the queue depth (for example, the maximum number of I/O operations that are
submitted in parallel). Depending on the type of host and type of host bus
adapters it may be possible to limit the queue depth per virtual disk and/or
limit the queue depth per host bus adapter. An alternative method of limiting
the I/O workload generated by a host would be to use the I/O governing features
provided by the IBM Total Storage SAN Volume Controller. These techniques may be particularly
applicable if using a mixture of different host systems to prevent one host
system from saturating an I/O subsystem to the detriment of the other host
systems. Note that although these techniques may be used to avoid I/O time-outs
it still means the performance of your system is being limited by the amount
of storage.