Profit
Based Two-Step Job Scheduling in Clouds
Shuo Zhang, Li Pan
(
✉
)
, Shijun Liu, Lei Wu, and Xiangxu Meng
School of Computer Science and Technology, Shandong Uiversity,
Jinan 250101, People’s Republic of China
zs_sduzz@sina.com, {panli,lsj,i_lily,mxx}@sdu.edu.cn
Abstract. One of the critical challenges facing the cloud computing industry
today is to increase the profitability of cloud services. In this paper, we deal with
the problem of scheduling parallelizable batch type jobs in commercial data
centers to maximize cloud providers’ profit. We propose a novel and efficient
two-step on-line scheduler. The
first
step is to rank the arrival jobs to decide an
eligible set based on their inherent profitability and pre-allocate resources to them;
and the second step is to re-allocate resources between the waiting jobs from the
eligible set, based on threshold
profit-effectiveness
ratio as a
cut-off
point, which
is decided dynamically by solving an aggregated revenue maximization problem.
The results of numerical experiments and simulations show that our approach are
efficient in scheduling parallelizable batch type jobs in clouds and our scheduler
can outperform other scheduling algorithms used for comparison based on clas‐
sical heuristics from literature.
Keywords: Cloud · Resource allocation · Scheduling · Profit maximization
1 Introduction
Cloud computing is now significantly changing the way people use resources such as
computation hardware, storage, software applications and so on. Users can obtain on-
demand cloud services to host their jobs and applications in a pay-per-use way. With
virtualization as a key enabler, cloud providers can run multiple isolated Virtual
Machines (VMs) simultaneously in their data centers to host users’ jobs. By maximizing
server consolidation, cloud providers can achieve efficient utilization of resources and
thus obtain profits through economy of scale [1].
In an open and dynamic cloud environment, users have divergent requirements on
services for their jobs. They usually need to pay more to get better Quality of Service
(QoS), while in order to deliver better services cloud providers need to provision more
resources for hosting jobs and thus incur more cost. More specifically, there exists
resource competition between users’ jobs submitted to cloud data centers. How to
schedule these concurrent service requests efficiently on the physical resources in cloud
data centers to maximize a cloud provider’s profit is far from trivial.
In this paper, as an example application, we consider scheduling parallelizable batch
type jobs into cloud data centers. One characteristic of these jobs is that their execution
can be speeded up to an extent by allocating more resources to them. Besides, for these
types of jobs, response time is a main QoS criterion. Under this circumstance, to solve
© Springer International Publishing Switzerland 2016
B. Cui et al. (Eds.): WAIM 2016, Part II, LNCS 9659, pp. 481–492, 2016.
DOI: 10.1007/978-3-319-39958-4_38
剩余11页未读，继续阅读
资源评论
