Information Technology Reference
In-Depth Information
amount of energy saving depends on the number of hosts that have the highest
performance-per-watts ratio. That is if the number of type C servers (i.e. those
with the highest TotalMIPS/Watts ratio) is decreased, then the energy saving
is also decreased.
6
Conclusion and Future Work
In this paper, the problem of VM allocation to reduce energy consumption while
satisfying the fulfillment of quality of service (e.g. performance and resource
availability on time user requested) in HPC Cloud is studied. We presented
EPOBF, which is a power-aware allocation heuristic of VMs in HPC Cloud,
and can be applied to HPC clouds. A HPC clouds scheduler can use the met-
ric of performance-per-watt to allocate VMs to hosts for more energy eciency.
The experimental simulations show that we can significantly reduce energy con-
sumption in comparison with the state-of-the-art power-aware allocation heuris-
tics (e.g. PABFD). The EPOBF-ST and EPOBF-FT heuristics could be a new
VM allocation solution in a Cloud data center with heterogeneous and multi-
core physical machines. Both versions of EPOBF-ST and EPOBF-FT heuristics
are better than the PABFD and VBP greedy L1/L2/L30 allocation heuristics.
The percentage of energy saving depends on how much energy consumption on
server types and number of hosts that have the highest performance-per-watts
ratio.
A limitation of our work is evaluating the performance of the EPOBF-ST
and EPOBF-FT heuristics on various system models and workloads to pro-
vide a pros and cons of the EPOBF-ST and EPOBF-FT heuristics. Further-
more, we will consider limitations on computing resources and the impact of
other components such as physical memory, network bandwidth in performance
and energy consumption. In future, we plan to integrate the EPOBF-ST and
EPOBF-FT heuristics into a cloud resource management software (e.g. Open-
Stack Nova Scheduler). The cloud systems can provide resources to many types of
VM-based leases [
22
] including best-effort, advanced reservation, and immediate
leases at the same time. We are also studying Mixed Integer Linear Program-
ming models and meta-heuristics (e.g. Genetic Algorithms) of the VM allocation
problem.
References
1. AWS - High Performance Computing - HPC Cloud Computing.
http://aws.
amazon.com/hpc/
(retrieved on August 31, 2014)
2. Parallel
Workloads
Archive.
(retrieved on January 31, 2014)
3. SDSC-BLUE-2000-4.1-cln.swf.gz log-trace.
http://www.cs.huji.ac.il/labs/parallel/
workload/l sdsc blue/SDSC-BLUE-2000-4.1-cln.swf.gz
(retrieved on Januray 31,
2014)
Search WWH ::
Custom Search