Introduction to the New Mainframe: Large-Scale Commercial Computing

40 Introduction to the New Mainframe: Large-Scale Commercial Computing

A more detailed discussion on what scalability means and why it is important can

be found in the paper entitled “Scaling—Up and Out”, which is available at the

following site:

http://www.redbooks.ibm.com/abstracts/redp0436.html

The workload management function is a software piece that plays an important

role in the management of these different workload types, and thus on the

scalability of a system.

3.2.2 Scalability influences

Scalability refers to the ability of a system to provide nondisruptive growth path in

order to handle the projected explosion in computing demand. Scalability

includes not only processor power, but also processor memory, I/O bandwidth,

the ability to attach to large number of peripherals, the ability of the software to

exploit large systems and the ability to manage a large system.

Almost all platforms today are positioned as scalable systems. The IBM System z

server family has the proven capability to support tens of thousands of users,

while processing hundreds or even thousands of transactions per second.

There are two broad approaches to processor scalability—vertical and horizontal

growth. These are explained in more detail here.

Vertical growth

Vertical growth is the ability to upgrade the installed server processor capacity to

a larger one within the same family, for example, by adding further processor

engines to a Symmetric Multiprocessor (SMP).

The IBM System z multiprocessing design is a mature technology with decades

of expertise behind it. In order to reduce interprocessor interference, IBM has

improved both the hardware design and the software algorithms of the operating

systems that support the IBM System z. This means that there is a greater

scalability in model upgrades for IBM System z than for other platforms. This

design also means that the IBM System z family of processors offers clear,

simple upgrade paths between models.

Horizontal growth through Parallel Sysplex

Horizontal growth is the ability to add processor capacity by adding more servers

in a cluster. The IBM System z unique Parallel Sysplex share-all, clustering

technology offers near-linear scalability and a capacity far exceeding any single

commercial workload, by allowing you to connect up to 32 IBM System z servers

and to operate and manage them as a single image.

Chapter 3. Scalability 41

Figure 3-1 Linear growth compared to Sublinear Performance

Figure 3-1 shows linear growth compared to sublinear performance. The

performance number is based on real-world test results. Nways® represents the

number of processors. By “sublinear” we mean that the real (measured)

performance growth is less than one would expect from drawing a straight line

that is defined by the first two measurement points (in mathematics, sublinear

means “less than linear”

4

).

In Figure 3-1 that line is also shown, so the nth CP of a Parallel Sysplex adds

less computing power to the system than the one before (n-1). In a linear growth

(top line), all CP would add the same amount of computing power.

Clusters, or scale out systems, replicate complete systems interconnected via a

variety of interconnect mechanisms. In scale out systems, the performance and

scalability are limited by both the speed and efficiency of the intra-node

communication and by the effectiveness of the sharing scheme implemented at

the software level.

There are many clustering schemes, depending on the machine type used. The

IBM System z cluster technology is called Parallel Sysplex, and is described in

detail Chapter 5, “Availability” on page 87.

4

From Wikipedia Encyclopedia http://en.wikipedia.org/wiki/Sublinear

IBM System z

0

5000

10000

15000

20000

25000

0

5000

10000

15000

20000

25000

1

3

5

7

9

11

13

15

17

19

21

23

25

27

29

31

Nways

Relative Performance

Sublinear

Performance

Sublinear

Performance

Linear

growth