Principles, Methodology and Tools for Engineering Cloud Computing Systems - Advanced Design Approaches to Emerging Software Systems - page 265

in such circumstances. These possibilities are
likely to inhibit the adoption of Cloud comput-
ing, and means of mitigating these risks though
clear contractual agreements between service and
infrastructure provider must exist. These agree-
ments should define the obligations of both parties
regarding the level of service to be provided and
the behavior deemed acceptable on both parts,
and enable financial compensation to be sought
should a party fail to meet these obligations.
We are primarily concerned here with agree-
ments between service provider and Cloud in-
frastructure provider regarding the performance
constraints of the physical resources allocated to
an overall application service. As such, the SLA
must define expected performance, reliability,
and the conditions within which these can be
guaranteed, in addition to the compensation to
pay when the agreed objectives are not matched.
Most Cloud providers will describe some form
of SLA. Amazon for example details a broad
commitment of 99.95% uptime for all service
instances and will provide some form of credit
should the performance not be met 7 . Users can
provide details by email regarding the observed
failures including date and time.
However, when dealing with large scale
services involving multiple components poten-
tially distributed between several locations and
administrative domains, such measures can be
found to be insufficient. The process of gathering
performance records and evaluating them against
expected detailed quality of service requirements
must be automated as much as possible. This re-
quires service level objectives to be specified in a
clear and unambiguous manner that can be evalu-
ated at run time against observed performance
measurements. In addition service providers may
have several concerns beyond projected uptime.
A failure to respond to elasticity requirements in
a timely manner may lead to an inability to meet
overall demand and generally the provider will
be concerned with performance level objectives
tied to application specific behavior, such as the

overall turn around time of individual service
requests involving multiple components.
Several standards and frameworks exist for the
definition, negotiation and monitoring of service
level agreements. Many of these are tied to specific
types of applications, but we may consider for
example the WS-Agreement specification as an
example of a highly extensible framework suited
for such a purpose (OGF, 2006).
Generally an SLA description will require
some form of service description, enabling us to
pinpoint the specific characteristics of the service
that the SLA is meant to protect, guarantee terms
in the form of specific service level objectives,
the conditions within which these apply and the
perceived business value of these objectives. In
addition, appropriate metrics used as a basis for
the formulation of the service level objectives will
have to be defined with respect to the application
domain, and a monitoring environment. It must be
possible to obtain and communicate the value of
key performance indicators specified in the SLA
via appropriate measurement probes.
An example of how SLAs may be monitored
and generally integrated within the rest of a Cloud
platform is illustrated in Figure 6. An SLA com-
pliance monitor is responsible for parsing a set
of guarantee terms and conditions from an SLA,
such as a WS-Agreement based document. This
can then evaluate dynamically compliance to
these rules by matching measurements obtained
from measurement probes at various levels of
the infrastructure to specific metrics specified
in the SLA. Conditional expressions can then be
evaluated against the collection of measurement
records obtained in specified time frames.
Identified violations can then be passed to
suited components, such as an SLA protection
engine, which would try to adjust allocations ac-
cordingly, or alternatively to a business informa-
tion manager which will enact some penalty in
the form of financial compensation or otherwise.
Such penalties would be described in the SLA.