Resource allocation for job optimization in multi-cloud environment
Resource management consists of three domains namely allocation, discovery, and monitoring. Resource allocation in cloud computing is a complex process that involves identifying the best pair of tasks and resources based on quality-of-service requirements. Hence, the agility of demands for job pr...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English |
| Published: |
2023
|
| Subjects: | |
| Online Access: | http://psasir.upm.edu.my/id/eprint/112700/1/112700.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Resource management consists of three domains namely allocation, discovery, and
monitoring. Resource allocation in cloud computing is a complex process that involves
identifying the best pair of tasks and resources based on quality-of-service requirements.
Hence, the agility of demands for job processing from the clients is a challenge for cloud
service broker to efficiently allocate resources and meet the requirements of the tasks
within the specified deadline.
This thesis studies the resource management problem in resource allocation of multi
cloud environment. The main problem is when the allocation of resources influences the
optimization of job processing and the cloud resources. It leads the resources to be
underutilized or overutilized, resulting in poor resource utilization and inefficient job
execution. This thesis analyses the current optimization solution, used preemption
mechanism via dynamic cloud list scheduling (DCLS) and dynamic min-min scheduling
(DCMMS) method. The current solution might cause higher execution time and lower
the utilization rate. Therefore, it is essential to provide an efficient mechanism for
resource allocation of job optimization to reduce the execution time and increase the
utilization time. The resource allocation and selection mechanisms are proposed for
cloud broker of job optimization in a multi-cloud environment. It also proposes high
level service brokering model to support the allocation and selection mechanisms. A
Multilevel Allocation mechanism (MLA) includes jobs and resources in allocation
mechanism as an effort to optimize job processing and resource allocation. The
allocation approach explicitly considers priority list and rank the resources for job
allocation. To leverage on the feedback information and processing power of a resource,
a Resource optimization Based on Reputation mechanism (REP-R) is being introduced.
The proposed mechanism deals with both job and resources simultaneously. Finally, a
selection mechanism method, Resource Selection Based on Job Classification, (RES-J)
is being proposed to select the fit resources based on job classification. Decision tree
classification is adopted for job classification; thus, it enables the discovery and
optimization of resource availability due to over or under provisioning.
To simulate the proposed mechanisms, CloudSim is used to conduct an extensive
simulation with a diverse set of jobs and scenarios. The findings of the mechanism show
that MLA is 80% better than other DAG methods. In producing shorter schedule length,
DCLS produces a better schedule length ratio (SLR) compared to the proposed
mechanisms of MLA and DCMMS. However, MLA is the best possible allocation of
resources or scheduling strategy to achieve the objective by minimizing the schedule
length since the mechanism considers many parameters as compared to DCLS when
scheduling the job. In contrast with SLR, MLA produces the best makespan among the
three mechanisms. For the second mechanism, the average execution time in REP-R is
7% faster than DCMMS and it outperforms DCLS. This is due to the allocation that
chooses the most reputable resources, thus minimizes the job execution time. For the
third mechanism, the overall performance shows that RES-J utilizes the most resources
compared to DCMMS and DCLS in booth loose and tight scenarios.
Overall, the proposed mechanism comprises multi-level, resource reputation and
selection in the allocation of resources, and it shows promising results in improving
resource utilization and overall performance of cloud systems. In addition, it is a strategy
for the cloud broker with the aim to minimize the overall cost and optimize job
scheduling. |
|---|