IBM eServer Cluster 1600 Managed by PSSP 3.5: What's New

Chapter 1. IBM eServer Cluster 1600 3

1.1 Clusters defined

A

cluster

is a set of two or more computers with some other unifying

characteristics. There are several different types of clusters. Let’s distinguish

among the types of clusters for the purposes of discussing the Cluster 1600.

Each of these clusters is more tightly clustered than the previous:

Islands: This is the cluster in your basement, that pile of PCs without Ethernet

cards that you used to play games on. A cluster? No.

Partitions: These can be log

ical

or

physical

partitions, or

virtual machines

, but

are all separate computers residing on the same piece of hardware. These

computers, all running their own operating system (OS) images and applications,

and having their own network addresses, may or may not be part of a cluster,

depending on how they are networked and managed.

򐂰 An example of physical partitioning is the planned support for growing the

xSeries x440 in 4-way stackable boxes from a 4-way all the way to 16-way

with a scalable non-uniform memory access (NUMA) interconnect between

the chunks. The system can be split up into physical partitions as well, each

running its own Linux or Windows OS.

򐂰 More flexible is the pSeries POWER4 systems (p690, p670, p655, p650),

which can be divided into separate AIX or Linux machines through logical

partitioning. A firmware hypervisor slices up the physical resources of the

server (I/O, CPU, and memory) so that multiple operating systems can be run

simultaneously. In fact, logical partitions can be allocated with fractional

central processing unit (CPU) and I/O resources. IBM mainframes have had

this kind of fractional logical partitioning for years with the PRSM technology.

pSeries currently offers logical partitioning on CPU and I/O slot boundaries,

but will offer fractional partitioning in the future.

򐂰 IBM invented virtual machines, of course, with the VM hypervisor for the

mainframe. Virtual machines are created by a software hypervisor that

manages special kernel-involved events, including privileged operation

interrupts, I/O interrupts, page faults, timers, and so on. These are dealt with

by the hypervisor, but the results are presented to the OS and software on the

virtual machines in a way identical to that presented by a real machine. CPU

resources are time-sliced by the hypervisor. Thus, the OS and software run

as if they had their own hardware. Because only kernel-involved events are

simulated, the large bulk of the operation of the virtual machine’s software

runs directly on the metal during its time-slice, offering good performance.

z/VM is still offered on the zSeries and is a popular choice for those wanting

to run multiple Linux images on these systems. VMWare ESX Server, in

partnership with IBM, offers similar virtualization for the PC server

architecture on many xSeries servers.