Page 226 - Cloud computing: From paradigm to operation

1 Framework and requirements for cloud computing

Table II.4 – Using an interconnect network for high-performance service – use case
Derived – Sharing process unit component (see clause 7.1.5)
requirements – Interconnect network supports (see clause 7.1.5)
– Device driver and API supports (see clause 7.2)
– Expansion of interconnect network (see clause 7.4)
– Network topology (see clause 7.1.5)

Table II.5 – Physical machine for hyper-scale deployment – use case

Title Physical machine for hyper-scale deployment
Description This use case covers the situation where very large numbers of physical machines will be
employed in multiple data centres around a region or around the world. In this case, a single
physical machine will typically be implemented as a server blade that fits into a specially built
rack. The rack will typically also include storage, management and networking equipment,
which might or might not be implemented using similar blades.

Roles/sub-roles CSP: cloud service operations manager.
Figure In this type of deployment, physical
machines are constructed for
deployment in high-density racks
specifically designed for the purpose.
Each rack provides all the
infrastructure required to support
the machines, including power,
network connectivity and
ventilation/cooling. Individual
physical machines are automatically
initialised and provisioned with
necessary software when plugged
into an active rack. Once running,
each machine is made available for
the deployment of VMs or other
functions requested by CSCs or CSP
administrators. Fault tolerance is
often provided by software across
multiple machines, so an individual
machine can fail without adverse
effect on the overall system.

Pre-conditions CSP: The CSP wishes to deploy many physical machines in two or more locations, using
(optional) minimal staffing.
Post-conditions A physical machine is plugged into a rack, configures itself to the local network and is
(optional) automatically provisioned with all necessary software including the host operating system,
network stacks and hypervisor. The data centre management system is then able to deploy
workloads to the machine. In the event of failure, the machine can be removed from the rack
and a replacement plugged in with minimal impact on deployed cloud services.
A data centre can be left unmanned for several days or weeks at a time. When visited, the
failed machines can be quickly removed and replaced by new or refurbished machines, which
configure and provision themselves automatically.

218