Evolution Of Data Centers; Density; Scale Out Applications - IBM NeXtScale System Planning And Implementation Manual

1.1 Evolution of data centers

There is an increasing number of computational workloads that can be run on
groups of servers; often referred to by such names as clusters, farms, or pools.
This type of computing can be described as scale-out, although, as a convention,
we refer to these groups as clusters. As the computing community's proficiency
with implementing and managing clusters improved, there is a trend to create
large clusters, which are becoming known as hyper-scale environments.
In the past, when the number of servers in a computing environment was lower,
to reduce application downtime, considerable hardware engineering effort and
server cost was expended to create servers that were highly reliable. With
clusters of servers, we strive to create a balance between the high availability
technologies that are built in to every server, and reduce the cost and complexity
of the servers, which allows more of them to be provisioned.

1.1.1 Density

As the number of servers in clusters grows and as data center real estate cost
increases, the number of servers in a unit of space (also known as the compute
density) becomes an increasingly important consideration. IBM NeXtScale
System is designed to optimize density while addressing other objectives, such
as, providing the best performing processors, minimizing the amount of energy
that is used to cool the servers, and providing a broad range of configuration
options.

1.1.2 Scale out applications

The following applications are among those that lend themselves to clusters of
servers:
High performance computing (HPC)
HPC is a general category of applications that are computationally complex,
can deal with large data sets, or consist of vast numbers of programs that
need to be run. Examples of computationally complex workloads include
weather modeling or simulating chemical reactions. Comparing gene
sequences is an example of a workload that involves large data sets. Image
rendering for animated movies and Monte Carlo analysis for particle physics
are examples of workloads where there are vast numbers of programs that
need to be run. The use of several HPC clusters in a Grid architecture is an
approach that gained popularity.
2
IBM NeXtScale System Planning and Implementation Guide