Filter
Top Products
Chapter 11. Performance considerations 237
With 15K RPM DDMs you need the equivalent of three 8 packs to satisfy the I/O load from the
host for this example. Note the DS6000 can also be configured with a RAID array comprised
of four DDMs.
Depending on the required capacity, you then decide the disk capacity, provided each
desired disk capacity has 15k RPM. When the access density is less and you need more
capacity, follow the example with higher capacity disks, which usually spin at a slower speed
like 10k RPM.
In “Fibre Channel device adapter” on page 226 we stated that the disk storage subsystem DA
port in a DS6000 has about five times more sequential throughput capability than an ESS 800
DA port provides. Based on the 2 Gbps Fibre Channel connectivity to a DS6000 disk array,
this is approximately 200 MB/sec compared to the SSA port of an ESS disk array with 40
MB/sec. A Fibre Channel RAID array provides an external transfer rate of over 200 MB/sec.
The sustained transfer rate varies. For a single disk drive various disk vendors provide the
following numbers:
146 GB DDM with 10K RPM delivers a sustained transfer rate between 38 and 68 MB/sec,
or 53 MB/sec on average.
73 GB DDM with 15K RPM transfers between 50 and 75 MB/sec, or 62.5 MB/sec on
average.
The 73 GB DDMs have about 18% more sequential capability than the 146 GB DDM, but
60% more random I/O potential. The I/O characteristic is another aspect to consider when
deciding the disk and disk array size. Note that this discussion takes a theoretical approach,
but it is sufficient to get a first impression.
At GA the IBM internal tool, Disk Magic, helps to model configurations based on customer
workload data. An IBM representative can contact support personnel who will use Disk Magic
to configure a DS6000 accordingly.
Use Capacity Magic to find out about usable disk capacity. This tool is also available at an
IBM internal intranet sales site.
11.5.4 Configuration recommendations for z/OS
We discuss briefly how to group ranks into extent pools and what the implications are with
different grouping approaches. Note the independence of LSSs from ranks. Because an LSS
is congruent with a z/OS LCU, we need to understand the implications. It is now possible to
have volumes within the very same LCU, which is the very same LSS, but these volumes
might reside in different ranks and the ranks might be on different loops.
A horizontal pooling approach assumes that volumes within a logical volume pool, like all DB2
volumes, are evenly spread across all ranks and loops. This is independent of how these
volumes are represented in LCUs. The following sections assume horizontal volume pooling
across ranks, which might be congruent with LCUs when mapping ranks accordingly to LSSs.
Configure one extent pool for each single rank
Figure 11-11 on page 234 displays some aspects regarding the disk subsystem within a
DS6000:
Chapter 4, “Virtualization concepts” on page 65, introduced the construct of an extent
pool. When defining an extent pool an affinity is created between this specific extent pool
and a server. Due to the virtualization of the Fibre Channel switched disk subsystem you
might consider creating as many extent pools as there are RAID ranks in the DS6000.
This would then work similar to what is currently in the ESS. With this approach you can