S.S. Manvi, G.

Krishna Shyam / Journal of Network and Computer Applications ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 7

ability to experience failures or systematic attacks, without impact- perform multi-tier dynamic resource provisioning. In particular,
ing customers or operations. We analyzed whether the proposed database tier scaling is not considered. Also, the authors do not
system (in different research papers) scales itself with increase in follow any approach toward dynamic resource management on
number of users. If the system does not complicate itself with clouds.
increase in number of users, it is said to be highly reliable. If the Inefficiency of resource provisioning leads to either overprovi-
system works with too much constraints, then it is not reliable. sioning or underprovisioning problem. Vijayakumar et al. (2010a)
Ease of deployment refers to the easiness in implementing the propose a robust cloud resource provisioning (RCRP) algorithm to
system model. The value for ease of deployment has been assigned minimize the total resource provisioning cost (i.e., overprovision-
as high if the infrastructures are easily available for deployment. ing and underprovisioning costs). Various types of uncertainty are
Quality of Service (QoS) refers to a broad collection of networking considered in the algorithm. Dailey et al. (2011) propose a method
technologies and techniques. The goal of QoS is to provide for identifying and retracting overprovisioned resources in multi-
guarantees on the ability of a network to deliver predictable tier cloud-hosted Web applications. They demonstrate the feasi-
results. The elements of network performance in our work which bility of approach in an experimental evaluation with a testbed
decides QoS includes availability (uptime), bandwidth (through- EUCALYPTUS based cloud and a synthetic workload. But the
put), latency (delay), and error rate. The higher availability, higher problem is that they only address scaling of the Web server tier
bandwidth, lower latency, and lower error rate offers higher QoS. and a read-only database tier. In particular, they do not address
Delay is the time taken from point-to-point in a network. Higher software configuration management.
delay degrades performance of the system and vice versa. Buyya et al. (2011) point out many challenges in addressing the
Control overhead refers to the extra consideration required by a problem of enabling SLA-oriented resource allocation in data
system that is not directly related to data. In this paper, we made centers to satisfy competing applications demand for computing
an analysis of whether any resources are consumed or lost in services. In particular, the user applications are becoming more
completing a process that does not contribute directly to the end- complex and need multiple services to execute instead of a single
product. If the resources are lost or extra resources required during service. In particular, work on cloud management in Cunningham
transmission of the data, it means that the control overhead and Holmes (2011) and Armbrust and Fox (2010) has focused on
is high. the provisioning and scaling of services within infrastructure
clouds. Among the problems that are faced by the users, perfor-
5.1. Resource provisioning mance and virtualization problems are the most persistent and
prevalent problems owing in part due to the fact that users have
The development of efficient service provisioning policies is no visibility into the cloud and are thus forced to consult the cloud
among the major issues in cloud research. The issue here is to operators for help. The surveys indicate that to offer more effective
provide better quality of service in IaaS by provisioning the support, clouds should develop tools to automate operators task.
resources to the users or applications via load balancing mechan- Vijayakumar et al. (2010b) have considered the problem of
ism, high availability mechanism, etc. In this context, game cost-sensitive resource provisioning for adaptive data streaming
theoretic methods in Teng and Magoules (2010) allows us to gain applications in virtualized or cloud environments. This framework
an in depth analytical understanding of the service provisioning dynamically achieves the user-specified accuracy level by adapting
problem. Earlier game theory has been successfully applied to a adaptive parameter at runtime. Chaisiri et al. (2012) have
diverse problems such as Internet pricing, congestion control, proposed an optimal cloud resource provisioning (OCRP) algo-
routing, and networking. Resource provisioning can encompass rithm to provision resources offered by multiple cloud providers.
three dimensions as per Sotomayor et al. (2009): hardware The optimal solution obtained from OCRP is found by formulating
resources, the software available on those resources, and the time and solving stochastic integer programming. The OCRP algorithm
during which those resources must be guaranteed to be available. can be used as a resource provisioning tool for the emerging cloud
A complete resource provisioning model must allow resource computing market in which the tool can effectively save the
consumers to specify requirements across these three dimensions, total cost.
and the resource provider to efficiently satisfy those requirements. Warneke and Kao (2011) have discussed the challenges and
The service provisioning procedure according to Hill and opportunities for efficient parallel data processing in cloud envir-
Varaiya (2009) is based on a solution of the problem of allocating onments and presented Nephele, the first data processing frame-
bandwidth and buffers to meet several types of service requests, work to exploit the dynamic resource provisioning offered by
differentiated by bounds on the average rate and burstiness of the today IaaS clouds. They have described Nephele basic architecture
message and on the end-to-end delay. Here, the users decide the and presented a performance comparison to the well-established
resources they need and the network coordinates their choices via data processing framework Hadoop. The performance evaluation
resource pricing in order to optimize an overall measure of gives a first impression on the ability to assign specific virtual
network performance. machine types to specific tasks of a processing job, as well as the
Among the existing works that we have at present for resource possibility to automatically allocate/deallocate virtual machines in
provisioning in cloud computing, we observed that only few the course of a job execution. This can help in improving overall
researchers have addressed the problem in multi-tier applications. resource utilization. and consequently reduce the processing cost.
The work given in Urgaonkar et al. (2010b) presents a model which Juve and Deelman (2012) have discussed several techniques
can be best described as a analytical model using queuing networks based on resource provisioning that may be used to reduce
in which the behaviour of each tier has been captured. This analytical network overheads. These techniques include: advance reserva-
model is able to predict parameters such as the think time, service tions, multi-level scheduling, and Infrastructure as a Service (IaaS).
time and visit ratio. The most recent work in this area given in Singh They have discussed the advantages and disadvantages of these
et al. (2010) which presents a technique to model dynamic work- techniques in terms of cost, performance and usability.
loads for multi-tier Web applications using k-means clustering. The In Huang et al. (2011), an architectural design of on-demand
method uses queuing theory to model the system reaction to the service for grid computing is proposed. A profile-based approach
workload and to identify the number of instances required for an to capture expert knowledge of scaling applications was proposed
Amazon EC2 cloud to perform well under a given workload. Although in which extra demanded resources can be more efficiently
this work does model system behavior on a per-tier basis, it does not provisioned as in Jie et al. (2011). In Kee and Kesselman (2011),

Please cite this article as: Manvi SS, Krishna Shyam G. Resource management for Infrastructure as a Service (IaaS) in cloud computing:
A survey. Journal of Network and Computer Applications (2013), http://dx.doi.org/10.1016/j.jnca.2013.10.004i