IBM TotalStorage DS6000 Series: Performance Monitoring and Tuning

358 IBM TotalStorage DS6000 Series: Performance Monitoring and Tuning

10.1 Overview

With its new and unique architecture, the DS6000 provides a high level of performance

across all the server platforms it attaches to.

Specifically for the zSeries servers, the DS6000 is a disk subsystem with a very good price

performance ratio. It should be able to handle sequential workload better than an ESS 800,

like data mining and work volumes. Large database applications do not work as well on the

DS6000, in which case you will need to use a DS8000.

The DS6000 features that have performance implications in the application I/O activity are

described in the following sections:

򐂰 Parallel Access Volumes

򐂰 Multiple Allegiance

򐂰 I/O Priority Queuing

򐂰 Logical volume sizes

򐂰 FICON

In the following sections of this chapter we describe these DS6000 features and discuss how

they can be used to boost the performance of your zSeries environment.

10.2 Parallel Access Volumes

Parallel Access Volume (PAV) is one of the original features that the ESS provides specifically

for the z/OS users. Simply stated, PAV allows multiple concurrent I/Os to the same volume at

the same time from applications running on the same z/OS system image. This concurrency

helps zSeries applications better share the same logical volumes with reduced contention.

The ability to send multiple concurrent I/O requests to the same volume nearly eliminates I/O

queuing in the operating system, thus reducing I/O responses times.

Traditionally, access to highly active volumes has involved manual tuning, splitting data across

multiple volumes, and more things in order to avoid those hot spots. With PAV and the z/OS

Workload Manager, you can now almost forget about manual device level performance tuning

or optimizers. The Workload Manager is able to automatically tune your PAV configuration

and adjust it to workload changes. The DS6000 in conjunction with z/OS has the ability to

meet the highest performance requirements.

PAV is implemented by defining alias addresses to the conventional base address. The alias

address provides the mechanism for z/OS to initiate parallel I/O to a volume. As its name

implies, an alias is just another address/UCB that can be used to access the volume defined

on the base address. An alias can only be associated with a base address defined in the

same LCU. The maximum number of addresses you can define in an LCU is 256.

Theoretically you can define 1 base address plus 255 aliases in an LCU.

10.2.1 Static and dynamic PAVs

Aliases are initially defined to be associated to a certain base address. In a static PAV

environment, the alias is always associated to the same base address, while in a dynamic

PAV environment, an alias can be reassigned to any base address as need dictates.

With dynamic PAV, you do not need to assign as many aliases in an LCU as compared to a

static PAV environment, because the aliases will be moved around to the base addresses that

need an extra alias to satisfy an I/O request.