IBM TotalStorage DS6000 Series: Performance Monitoring and Tuning

360 IBM TotalStorage DS6000 Series: Performance Monitoring and Tuning

Before the ESS, a device had an implicit allegiance, that is, a relationship created in the disk

control unit between the device and a channel path group, when an I/O operation was

accepted by the device. The allegiance caused the control unit to guarantee access (no busy

status presented) to the device for the remainder of the channel program over the set of paths

associated with the allegiance.

With Multiple Allegiance (MA), the requests are accepted by the DS6000 and all requests will

be processed in parallel, unless there is a conflict when writing data to the same extent of the

CKD logical volume. Still, good application access patterns can improve the global parallelism

by avoiding reserves, limiting the extent scope to a minimum, and setting an appropriate file

mask, for example, if no write is intended.

In systems without Multiple Allegiance, all except the first I/O request to a shared volume are

rejected, and the I/Os are queued in the zSeries channel subsystem, showing up as PEND

time in the RMF reports.

Multiple Allegiance provides significant benefits for environments running a Sysplex or

zSeries systems sharing access to volumes. Multiple Allegiance and PAV can operate

together to handle multiple requests from multiple hosts.

The DS6000 ability to run channel programs to the same device in parallel can dramatically

reduce the IOSQ and the PEND time components in shared environments.

In particular, different workloads—for example, batch and online—running in parallel on

different systems can have an unfavorable impact on each other. In such cases, Multiple

Allegiance can dramatically improve the overall throughput.

10.4 How PAV and Multiple Allegiance work

These two functions allow multiple I/Os to be executed concurrently against the same volume

in a z/OS environment. In the case of PAV, the I/Os are coming from the same LPAR or z/OS

system, while for Multiple Allegiance, the I/Os are coming from different LPARs or z/OS

systems.

First we will look at a disk subsystem that does not support both of these functions. If there is

an outstanding I/O operation to a volume, all subsequent I/Os will have to wait as illustrated in

Figure 10-1 on page 361. I/Os coming from the same LPAR will wait in the LPAR and this wait

time is recorded in IOSQ Time. I/Os coming from different LPARs will wait in the disk control

unit and be recorded in Device Busy Delay Time, which is part of PEND Time.

In the ESS and DS6000, all these I/Os will be executed concurrently using PAV and Multiple

Allegiance, as shown in Figure 10-2 on page 361. I/O from the same LPAR will be executed

concurrently using UCB

1FF that is an alias of base address 100. I/O from a different LPAR

will be accepted by the disk control unit and executed concurrently. All these I/O operations

will be satisfied from either the cache or one of the DDMs on a Rank where the volume

resides.