Information Technology Reference
In-Depth Information
are simplified by centralized web access interfaces to the cloud infrastructure.
Paremus and Jadex pursue the idea of a dynamic cloud with a varying number
of nodes. For this reason both use a daemon approach, meaning that a minimal
bootstrapping software has to be installed on all nodes of the cloud. The cloud
structure is rather well known for a typical cloud PAAS like Google App Engine
and CloudBees and made fully transparent to the cloud user. Instead, Paremus
and JadexCloud are built to deal with dynamic cloud infrastructures allowing
nodes to be discovered and dynamically included or excluded to/from the cloud.
Thus, the application view is different as well. The first two keep the standard
non-distributed application view and use internal logic for scaling, the latter two
handle truly distributed applications, meaning that functionally different parts
of applications run on different machines. Despite the application distribution
both, Paremus and JadexCloud, aim at a high-level view on the distributed ap-
plication being able to abstract from distribution aspects. All four platforms
deal with dynamic reconfiguration of applications. In the Google App Engine
and CloudBees, reconfiguration tasks are completely transparent. Paremus and
JadexCloud have to deal with more complex situations due to the possibly fine-
grained application deployment structure. Relocations of components are di-
cult to achieve in Paremus due to its reliance on SCA. In JadexCloud, relocations
can be achieved, as active components support dynamic service bindings.
Summarizing, most grid toolkits are built for dynamic environments and allow
for runtime grid adaptations but suffer from too simple programming models (ex-
cept Proactive) based on object orientation or services. On the contrary, cloud
PAAS infrastructures are typically built to make standard (web) applications
scalable and are narrowly focused (except Paremus). They consider applications
as being non-distributed and use established programming models like object
orientation and components. Approaches like Paremus and JadexCloud go be-
yond the typical cloud PAAS and facilitate building distributed systems running
in a dynamic infrastructure.
6 Summary and Outlook
This paper has argued that enterprises can benefit from private clouds to per-
form heavy computational tasks allowing a smooth scaling according to current
demands. Such a private cloud may not only consist of dedicated servers but may
also include normal computers used for daily work. In order to run applications
in such a dynamic cloud it is necessary to a) create a robust infrastructure from
a varying set of nodes and b) use a programming model allowing applications
being assembled of independent components that are dynamically coordinated.
As a solution the JadexCloud infrastructure has been presented, consisting of
three layers. The daemon layer is responsible for managing basic cloud resources
by e.g. announcing nodes participating in the network. The platform layer is used
for application management tasks such as deployment and starting of application
components. The application layer supports application development through
APIs and tools. Using a workflow management application example it has been
shown how a distributed cloud application can be built.
