Advanced Reservation (AR) is used to guarantee resource provisioning for many different types of
applications including workflows. This technique is still under a huge controversy in both Business and
Research communities because of its potentialit
y of reducing resource utilization. Most of the works
proposed in this domain suggest reservation for the whole workflow schedule, and on all available
resources at the same time, which worsen the problem of resource utilization.
Many solutions are introduced to improve resource utilization under advanced reservation through
generating relaxed and elastic reservation plans that local scheduling systems could modify to improve
utilization and decrease internal fragmentation. These solutions depend mainly on changing rigid AR,
which considered to be the most difficult kind of reservation, into relaxed and elastic ones through adding
extra time on the resulted schedule and then distributing it on all tasks of the workflow.
This paper presents a new autonomic algorithm (EWARP) for producing elastic reservation plans for
workflow applications which doesn’t add extra times. Instead, it depends on exploiting the timing gaps
produced by the different scheduling algorithms. The new algorithm use the technique of discovering
timing gaps, but modifies it, and adds to it to be used for producing an elastic reservation plan for
workflows. The results presented in this paper demonstrate how the proposed algorithm outperforms
existing works in the fields by a lower bound approximating 25%.This shows that (EWARP) algorithm
offer efficient and practical solutions for the problem of scheduling workflow applications under QoS
constrains.
The study is researching the fault tolerance in the large distributed
environments such as grid computing and clusters of computers in
order to find the most effective ways to deal with the errors
associated with the crash one of the devices in th
e environment or
network disconnection to ensure the continuity of the application in
the presence of the faults.In this paper we study a model of the
distributed environment and the parallel applications within it. Then
we provide a checkpoint mechanism that will enable us to ensure
continuity of the work used by a virtual representation of the
application (macro dataflow) and suitable for the applications
which uses work stealing algorithm to distribute the tasks which
are implemented in heterogeneous and dynamic environment.
This mechanism will add a simple cost to the cost of parallel
execution as a result of keeping part of the work during fault-free
execution. The study also provides a mathematical model to
calculate the time complexity i.e. the cost of this proposed
mechanism.