6. Conclusions
A significant research problem in cloud computing is finding a
tradeoff between power efficiency while maintaining high perfor-
mance efficiency. In this paper, we provide a detailed case study
using various workloads to highlight the inefficient power work-
flow scheduling in Hadoop. We exploit the concept of building
profiles for applications with certain workloads executing in small-
to medium-scale data centers. A profile-based energy-efficient
framework is proposed with a novel scheduler that makes a good
tradeoff among various factors consisting of the cost of VM place-
ment, power usage, CPU utilization, and the load factor that affect
the efficiency of a data center. The performance of the proposed
scheduler is compared to RTC and HEFT schedulers extensively.
We demonstrated through extensive simulation studies that our
proposed scheduling approach is at least as good as the other two
algorithms in all scenarios, and better in many cases. Results show
that the proposed scheduler dominates the benchmarked sched-
ulers in energy efficiency and exploits the data center resources by
increasing the multitenancy of VMs per PM.
The VM placement strategy utilized in this study allows the
proposed scheduler to maximize the number of idle machines in
the cluster. For smaller and mid-sized workloads, the proposed
scheduler is 19% and 38% more energy efficient than RTC and
HEFT; however, for very large workloads, the energy efficiency is
comparatively slightly better with 3% and 7%, respectively. These
results suit well for the low variability and high certainty of work-
loads suitable for small to medium scale data centers, however the
proposed scheduler does not fare well for large data centers with
uncertain workloads. Furthermore, this study limits the power
aware scheduling of workflows to CPU intensive workloads only.
At the moment, the proposed approach considers CPU utilization
towards the computation of energy costs of a workflow.
As a future direction, we plan to improve the accuracy in esti-
mating the expected runtime for workflow tasks within the APs.
We intend to model and include the cost of relevant resources
such as Storage I/O, network I/O with in the Application profile
for a holistic overview of energy efficiency in the data center.
Furthermore, we intend to investigate the security and privacy
implications highlighted by the multifarious users on various VMs
in a multitenant server.
Do'stlaringiz bilan baham: |