Hybrid Spot Instance based Resource Provisioning
Strategy in Dynamic Cloud Environment
Naidila Sadashiv Dilip Kumar S M R S Goudar
Redknee
Bangalore, India 560 045
Email: rsgoudarl@gmail.com
Dept. of Computer Sci. and Engg. Dept. of Computer Sci. and Engg.
Acharya Institute of Technology
Bangalore, India 560 107
Email: sadashiv@acharya.ac.in
University Visvesvarya College of Engg.
Bangalore, India 560 001
Email: dilipkumarsm@gmail.com
Abstrct-Utilization of resources to the maximum extent in
large scale distributed cloud environment is a major challenge
due to the nature of cloud. Spot Instances in the Amazon Elastic
Compute Cloud (EC2) are provisioned based on highest bid with
no guarantee of task completion but incurs the overhead of
longer task execution time and price. The paper demonstrates
the last partial hour and cost overhead that can be avoided
by the proposed strategy of Hybrid Spot Instance. It aims to
provide reliable service to the ongoing task so as to complete the
execution without abruptly interrupting the long running tasks
by redefning the bid price. The strategy also considers that on
demand resource services can be acquired when spot price crosses
on-demand price and thereby availing high reliability. This will
overcome the overhead involved during checkpointing, restarting
and workload migration as in the existing system, leading to
efcient resources usage for both the providers and users. Service
providers revenue is carefully optimized by eliminating the free
issue of last partial hour which is a taxing factor for the provider.
Simulation carried out based on real time price of various
instances considering heterogenous applications shows that the
number of out-of-bid scenarios can be reduced largely which
leads to the increased number of task completion. Checkpointing
is also minimized maximally due to which the overhead associated
with it is reduced. This resource provisioning strategy aims to
provide preference to existing customers and the task which are
nearing the execution completion.
Index Terms-Cloud Computing; Resource Provisioning; Spot
Instances; Bidding; Checkpointing; Reliability;
I. INTRODUCTION
Cloud computing is an emerging paradigm that has rev
olutionized the consumption model of resources. With this
paradigm, customers are able to access resources and services
on the fy based on a pay-as-you-go model. Resources can
also be booked in advance in order to be assured of the
availability of desired resources. This is similar to market
oriented system where maximizing the proft is the objective.
Due to the dynamic nature of cloud computing where in the
demand and supply of resource is diffcult to predict, many of
the resource are left idle. In order to utilize the idle resources
efciently, Spot Instance (SI) is introduced by Amazon [1].
Amazon is the pioneer in the introduction of the Spot
Instance. The aim of SI is to generate revenue from the under
utilized resources and there by consider the economy aware
applications. Users who seek the service make a bid based
on the current SI price that is publicly available for diferent
978-1-4799-5958-7/14/$31.00 ©2014 IEEE
type of instances. These instances are launched using the
highest bid strategy. If the new bidding price is more than
the current spot price, then the current user's Spot Instances
will be terminated in no time and is termed as out-of-bid.
Hence the pricing of these resources are volatile and depends
on the supply and demand in the market. In such a dynamic
pricing environment which is prone to out-of-bid situation
shall involve fault tolerant techniques. Most commonly used
fault tolerant techniques are migration, job duplication and
checkpointing-based approach. They involve in saving the state
of an application or a process during execution and restoring
the saved state, for which a large storage capacity and time
is required [6], [14], [15], [17]. Fault tolerance is achieved
by these approaches but at a very high cost which should
be shared by the providers as well as users [3], [16]. Users
are attracted to the minimum cost of the resource but end up
paying more, and sometimes it exceeds the on-demand price.
Varying types of applications like economy aware applications,
compute intensive, as well in data analytic scenarios, such as
execution of MapReduce tasks [9], [14] and HTC are now
aiming to reap the benefts of SI. Hence providers should
provide more fexibility such that users are given a chance
for their task completion rather than abrupt task termination
[11], [19]. Users owe these resources at nominal price but not
for free of cost.
Our main objective is to efciently utilize the idle resources
by avoiding the overhead during checkpointing and out-of
bid time. In this paper we propose a Hybrid Spot Instance
(HSI) resource provisioning strategy that gives a chance to
the existing users to continue with resource usage at out-of
bid time by redefning the running task's bid. We also present
the overhead that the service provider faces which will be
addressed by using the HSI. User's task will be migrated to
on-demand scheme when SI price meets the on-demand price.
The rest of this paper is organized as follows. In Section
II we discuss some of the recent related works and highlight
their limitations and drawbacks. In Section III we present the
Hybrid Spot Instance algorithm. In Section IV we show the
simulation results with discussion and in Section V we present
the concluding remarks.