Information Technology Reference
In-Depth Information
The underlying task distribution model with a rather centralized application
has inspired several grid approaches like BOINC
4
and GridGain
5
.Otherap-
proaches like the GlobusToolkit [3] and Proactive [1] have extended the com-
putational model from client/server to generic distributed applications. The
programming model of the approaches directly reflects their application shape
assumptions. While GridGain and Boinc use traditional object oriented tech-
niques with tasks as primary abstraction for work distribution, Globus envi-
sions a service oriented world consisting of introspectable and transient grid
services, which will be created and terminated on demand. Proactive proposes
using a model of components and active objects resembling very much the au-
thor's active component approach [7]. Regarding installation and administration
complexity GridGain and BOINC address an easy integration of nodes offering
installer-based solutions, whereas Globus and Proactive assume that an infras-
tructure model is explicitly set up, describing e.g. where registries are located and
components should be deployed. The dynamics of environments is addressed by
very different means. Globus uses service registries, GridGain uses dynamic node
discovery based on awareness and BOINC uses a centralized server infrastruc-
ture. Proactive is rather focused on the static infrastructure model. Dynamic
reconfiguration is supported by all approaches to some extent. GridGain and
BOINC allow dynamic distribution of tasks taking into account the current grid
structure. Globus allows dynamic binding of services by registry lookups and
Proactive components can also be rebound by stopping and restarting them.
Cloud PAAS centers on ecient execution of specific application types, cur-
rently dominated by web applications. Typically, cloud PAAS hides distribution
and concurrency aspects from developers and enable them to deploy a standard
application in the cloud. The application can be scaled by the cloud infras-
tructure according to the customer demands. The approach is appealing but
bounded by the narrow focus of existing PAAS infrastructures. Furthermore, to-
day's applications have to follow vendor specific APIs for characteristics that are
subject of scaling, e.g. the storage system. This easily leads to vendor lock-ins
and problems in case scaling does not work as expected.
Google App Engine
6
and Run@Cloud from CloudBees
7
are two typical
platforms in the direction described above. The first facilitates development of
standard web applications while the latter supports Java EE enterprise appli-
cations. In contrast Paremus [5] is the only platform similar to JadexCloud
targeted at general distributed applications for private clouds. The underlying
programming models are object orientation in case of the Google App Engine,
component orientation in case of Run@Cloud, and component service orienta-
tion (SCA) in case of Paremus. JadexCloud further advances the programming
model to active components, which rely on a structural model very similar to
SCA. Installation and administration requirements for the first two platforms
4
http://boinc.berkeley.edu/
5
http://www.gridgain.com/
6
http://code.google.com/intl/de-DE/appengine/
7
http://www.cloudbees.com/run.cb
