98
clearly demarcate between legal and technical issues. Additionally, the
contract and the SLA create a complete agreement despite residing in
different documents, so that it is important to make sure that changes do
not lead to contradictions in terms. The following sections do not take this
possible split into account and look at the topics in a holistic way.
Availability Issues
One important aspect of Cloud computing, and in many cases even the most
important one, are availability issues. Aspects of data availability, system
availability and system security issues. This chapter focuses on data
availability issues.
As availability is typically expressed as an availability percentage within a
contractually agreed computation period, the customer's objective would be
to assure 100% availability. However, this is impossible and also unrealistic
because, viewed statistically, the customer's infrastructure could also fail.
Having said this, the assurances above 90% are widespread. But the
customer must be aware of the fact that values of this kind can mainly only
be achieved by two measures: as widespread a distribution of the data as
possible and/or measurement only up to the borders of the datacentre.
Other options are the ability to connect datacentres to the Internet by
multiple lines, which are preferably operated by multiple carriers. This not
only mitigates a local connection failure but also a failure of the provider's
entire network.
44
However, the customer needs to take into account that he will be unable to
benefit from all of these measures if he only has a single, failure‐prone
connection himself.
Besides defining the data availability it is also advisable to contractually
agree the period within which data availability must be restored in the case
of failure. The periods may vary, but as a rule it will be important to ensure
that availability is restored quickly. In this context Backups are usually not
stored locally along with the original set of data but at a different
datacentre. This strategy allows for short‐term restoration of the data and
ensures availability even in the case of total failure of the datacentre with
the original data. In other words, this is a security gain compared to what
was previously a single proprietary enterprise datacentre.
44
An expanded version of this chapter can be found under