Principles, Methodology and Tools for Engineering Cloud Computing Systems - Advanced Design Approaches to Emerging Software Systems - page 268

scalability, etc. new elements are required to make
these tasks more transparent and automatic for
application providers.
Figure 7 shows a summary of the major features
with regard to provisioning that the different Cloud
layers expose. Every layer increases the offered
abstraction level; unfortunately, none of them
seems suitable enough for application providers
to focus on their business only.

et al., 2010). Thus, mechanisms to allow resources
to grow or shrink in accordance with their utili-
zation are required. Scalability is a gigantic task
that needs to be tackled at the different abstrac-
tion levels provided by the Cloud paradigm (see
Figure 8).
At the IaaS level, Amazon provides higher
level services, such as Amazon's Cloud Watch
and AutoScale (Varia, 2009) to automate the scal-
ing process of Amazon-deployed VMs. Similarly,
RightScale allows for VM replication according
to queues or user-defined hardware and process
load metrics (Rightscale 2010). However, scal-
ability can only be defined in terms of the variables
monitored in the server templates RightScale
provides. In other words, service-level metrics,
those really relevant for service providers, cannot
be employed. Other relevant example is Micro-
soft's Azure, the number of instances for Azure
is specified in an XML configuration file which
has to be manually changed so that Azure's the
fabric controller will automatically adjust the
number of running instances. These very same

2.4 Service Scalability

Provisioning the service is just part of the whole
picture. Controlling how the service (not just
VMs) can automatically be scaled is also a very
desirable feature to be asked for to current IaaS
Cloud providers (Rodero-Merino et al., 2010;
Cáceres et al., 2010). This section shows how
current methods and tools are limited for the task
of providing different degrees of scalability at the
service level.
Cloud computing is, partially, but importantly
about on-demand provision of resources (Cáceres

Figure 7. Elements offered for provisioning different components at common Cloud layers. IaaS offer
VM provisioning interfaces; PaaS development interfaces for provisioning software bundles (often
Web-based, e.g. servlets); SaaS offer elementary service management interfaces (e.g. interfaces for
publishing, provisioning