A Stochastic Model to Investigate Data Center Performance and QoS in IaaS Cloud Computing Systems

A Stochastic Model to Investigate Data Center Performance and QoS in IaaS Cloud Computing Systems

Cloud data center management is a key problem due to the numerous and heterogeneous strategies that can be applied, ranging from the VM placement to the federation with other clouds. Performance evaluation of cloud computing infrastructures is required to predict and quantify the cost-benefit of a strategy portfolio and the corresponding quality of service (QoS) experienced byusers. Such analyses are not feasible by simulation or on-the-field experimentation, due to the great number of parameters that have to be investigated. In this paper, we present an analytical model, based on stochastic reward nets (SRNs), that is both scalable to model systems composed of thousands of resources and flexible to represent different policies and cloud-specific strategies. Several performance metrics are defined and evaluated to analyze the behavior of a cloud data center: utilization, availability, waiting time, and responsiveness. A resiliency analysis is also provided to take into account load bursts. Finally, a general approach is presented that,starting from the concept of system capacity, can help system managers to opportunely set the data center parameters under different working conditions.

PROJECT OUTPUT VIDEO: (Click the below link to see the project output video):

 In order to integrate business requirements and application level needs, in terms of Quality of Service (QoS), cloud service provisioning is regulated by Service Level Agreements (SLAs): contracts between clients and providers that express the price for a service, the QoS levels required during the service provisioning, and the penalties associated with the SLA violations. In such a context, performance evaluation plays a key role allowing system managers to evaluate the effects of different resource management strategies on the data center functioning and to predict the corresponding costs/benefits.
 Cloud systems differ from traditional distributed systems. First of all, they are characterized by a very large number of resources that can span different administrative domains. Moreover, the high level of resource abstraction allows to implement particular resource management techniques such as VM multiplexing or VM live migration that, even if transparent to final users, have to be considered in the design of performance models in order to accurately understand the system behavior. Finally, different clouds, belonging to the same or to different organizations, can dynamically join each other to achieve a common goal, usually represented by the optimization of resources utilization. This mechanism, referred to as cloud federation, allows to provide and release resources on demand thus providing elastic capabilities to the whole infrastructure.

 On-the-field experiments are mainly focused on the offered QoS, they are based on a black box approach that makes difficult to correlate obtained data to the internal resource management strategies implemented by the system provider.
 Simulation does not allow to conduct comprehensive analyses of the system performance due to the great number of parameters that have to be investigated.

 In this paper, we present a stochastic model, based on Stochastic Reward Nets (SRNs), that exhibits the above mentioned features allowing to capture the key concepts of an IaaS cloud system.
 The proposed model is scalable enough to represent systems composed of thousands of resources and it makes possible to represent both physical and virtual resources exploiting cloud specific concepts such as the infrastructure elasticity. With respect to the existing literature, the innovative aspect of the present work is that a generic and comprehensive view of a cloud system is presented.
 Low level details, such as VM multiplexing, are easily integrated with cloud based actions such as federation, allowing to investigate different mixed strategies. An exhaustive set of performance metrics are defined regarding both the system provider (e.g., utilization) and the final users (e.g., responsiveness).

 To provide a fair comparison among different resource management strategies, also taking into account the system elasticity, a performance evaluation approach is described.
 Such an approach, based on the concept of system capacity, presents a holistic view of a cloud system and it allows system managers to study the better solution with respect to an established goal and to opportunely set the system parameters.


A Stochastic Model to Investigate Data Center Performance and QoS in IaaS Cloud Computing Systems

1. System Queuing
2. Scheduling Module
3. VM Placement Module
4. Federation Module
5. Arrival Process

1. System Queuing:
Job requests (in terms of VM instantiation requests) are en-queued in the system queue. Such a queue has a finite size Q, once its limit is reached further requests are rejected. The system queue is managed according to a FIFO scheduling policy.
2. Scheduling Module:
When a resource is available a job is accepted and the corresponding VM is instantiated. We assume that the instantiation time is negligible and that the service time (i.e., the time needed to execute a job) is exponentially distributed with mean1/μ.

3. VM Placement:
According to the VM multiplexing technique the cloud system can provide a number M of logical resources greater than N. In this case, multiple VMs can be allocated in the same physical machine (PM), e.g., a core in a multicore architecture. Multiple VMs sharing the same PM can incur in a reduction of the performance mainly due to I/O interference between VMs.

4. Federation Module:
Cloud federation allows the system to use, in particular situations, the resources offered by other public cloud systems through a sharing and paying model. In this way, elastic capabilities can be exploited in order to respond to particular load conditions. Job requests can be redirected to other clouds by transferring the corresponding VM disk images through the network.
5. Arrival Process:
Finally, we respect to the arrival process we will investigate three different scenarios. In the first one (Constant arrival process) we assume the arrival process be a homogeneous Poisson process with rate λ. However, large scale distributed systems with thousands of users, such as cloud systems, could exhibit self-similarity/long-range dependence with respect to the arrival process. The last scenario (Bursty arrival process) takes into account the presence of a burst whit fixed and short duration and it will be used in order to investigate the system resiliency

 System : Pentium IV 2.4 GHz.
 Hard Disk : 40 GB.
 Floppy Drive : 1.44 Mb.
 Monitor : 15 VGA Colour.
 Mouse : Logitech.
 Ram : 512 Mb.


 Operating system : Windows XP/7.
 Coding Language : JAVA/J2EE
 IDE : Netbeans 7.4
 Database : MYSQL

Dario Bruneo ,“A Stochastic Model to Investigate Data Center Performance and QoS in IaaS Cloud Computing Systems”,VOL. 25, NO. 3, MARCH 2014.

About the Author