IBM System storage DS6000 Series, Redbooks

Page: 255 / 406

Chapter 11. Performance considerations
231
11.4.2 Data placement in the DS6000
Once you have determined the disk subsystem throughput, the disk space and number of
disks required by your different hosts and applications, you have to make a decision regarding
the data placement.
As is common for data placement and to optimize the DS6000 resources utilization, you
should:
򐂰 Equally spread the LUNs across the DS6000 servers.
Spreading the LUNs equally on rank group 0 and 1 will balance the load across the
DS6000 servers.
򐂰
Use as many disks as possible.
򐂰 Distribute across DA pairs and loops.
򐂰 Stripe your logical volume across several ranks.
򐂰 Consider placing specific database objects (such as logs) on different ranks.
All disks in the storage subsystem should have roughly the equivalent utilization. Any disk that
is used more than the other disks will become a bottleneck to performance. A practical
method is to make extensive use of volume level striping across disk drives.
11.4.3 LVM striping
Striping is a technique for spreading the data in a logical volume across several disk drives in
such a way that the I/O capacity of the disk drives can be used in parallel to access data on
the logical volume. The primary objective of striping is very high performance reading and
writing of large sequential files, but there are also benefits for random access.
DS6000 logical volumes are composed of extents. An extent pool is a logical construct to
manage a set of extents. One or more ranks with the same attributes can be assigned to an
extent pool. One rank can be assigned to only one extent pool. To create the logical volume,
extents from one extent pool are concatenated. If an extent pool is made up of several ranks,
a LUN can potentially have extents on different ranks and so be spread over those ranks.
However, to be able to create very large logical volumes, you must consider having extent
pools that span more than one rank. In this case, you will not control the position of the LUNs
and this may lead to an unbalanced implementation as shown in Figure 11-10 on page 232.
Combining extent pools made up of one rank and then LVM striping over LUNs created on
each extent pool, will offer a balanced method to evenly spread data across the DS6000 as
shown in Figure 11-10.
Note: Database logging usually consists of sequences of synchronous sequential writes.
Log archiving functions (copying an active log to an archived space) also tend to consist of
simple sequential read and write sequences. You should consider isolating log files on
separate arrays.
Note: We recommend assigning one rank per extent pool to control the placement of the
data. When creating a logical volume in an extent pool made up of several ranks, the
extents for this logical volume are taken from the same rank if possible.