Cloud applications have been gaining popularity in recent years for their flexibility in resource provisioning according to Web application demands. The Elastic Application Container (EAC) system is a technology that delivers a lightweight virtual resource unit for better resource efficiency and more scalable Web applications in the Cloud. It allows multiple application providers to concurrently run their Web applications on this technology without worrying the demand change of their Web applications. This is because the EAC system constantly monitors the resource usage of all hosting Web applications and automatically reacts to the resource usage change of Web applications (i.e. it automatically handles resource provisioning of the Web applications, such as scaling of the Web applications according to the demand). In the chapter, the authors firstly describe the architecture, its components of the EAC system, in order to give a brief overview of technologies involved in the system. They then present and explain resource-provisioning algorithms and techniques used in the EAC system for demand-driven Web applications. The resource-provisioning algorithms are presented, discussed, and evaluated so as to give readers a clear picture of resource-provisioning algorithms in the EAC system. Finally, the authors compare this EAC system technology with other Cloud technologies in terms of flexibility and resource efficiency.
TopIntroduction
Driven by the rapid growth of the demand for efficient and economical computational power, cloud computing (Zhang, Cheng, & Boutaba, 2010), has led the world into a new era. By enabling virtualisation technology on physical machines (PMs), it not only gives immense benefits in terms of reliability, efficiency, and scalability, but also provides virtual computational services, such as computing power, storage and network, in such a way cloud users are able to consume them over the Internet as utilities.
Most notable cloud providers, such as, Amazon EC2 (Amazon, 2010), RightScale (Adler, 2011), offer Virtual Machines (VMs) as a service to cloud users and allow the users to host their web applications on the VMs. We refer this type of approach as VM + web applications approach as shown in Figure 1(a). It allows cloud users to directly control its underlying computing resources, such as VM operations, scaling, networking, etc. In addition, this approach allows the resources of a single PM to be shared across multiple VMs for maximum efficiency. However, setting up and maintaining a working environment for web applications are complex and time consuming for cloud users, and VM resource management is a heavy-weight task for the cloud providers in this approach. In practice, we have identified two scenarios showing VM + web applications approach less feasible and less resource-efficient.
Figure 1. (a) Architecture of VM + web application approach (b) Architecture of server + web application approach
Heavyweight VM Migration: VM migration over LAN (Local Area Network) is one of the most common VM resource management operations for cloud providers. However, the VM migration over LAN is a heavyweight task. In a shared-storage environment, a VM live migration requires transferring the working state and memory from one PM to another over LAN. It consumes a large amount of I/O and network traffics in the LAN environment (He, Guo, & Guo, 2011). In a WAN (Wide Area Network) environment, mechanisms for migrating VMs remain elusive. The VM-based migration across IDCs over the Internet (Wood, Ramakrishnan, van der Merwe, & Shenoy, 2010) also requires a huge amount of I/O for both IDCs and costs a great amount of time for replicating a VM from one IDC to another. This is because a VM generally consists of a guest OS and applications in which the size of the VM can vary from hundred megabytes to 50 gigabytes or more depending on the sizes of the OS and web applications as shown in Figure 1(a). This makes the VM resource management in WAN infeasible.
Overhead in the VM + Web application Approach: A VM generally consists of a guest operating system (OS) for running applications. The guest OS in the VM always occupies a considerable amount of VM resources, this causes the overhead in a VM. As number of VMs increases, the total amount of overhead in VMs increases in the cloud system. Therefore, the overhead in VMs consumes a great amount of cloud resources. This leads to resource-inefficiency in this approach.
In contrast, cloud providers, such as Google App Engine (Zahariev, 2009), OpenShift (Schabell), offer application servers to cloud users and allow cloud users to directly manage their web applications without considering the underlying computer resource consumption. We refer this type of approach as server + web applications approach as shown in Figure 1(b). The resource management of this approach becomes easier and more flexible than the VM + web applications approach. Scaling a web application is no longer based on VMs and it is based on each application instance of the web application. In addition, migrating a web application from one machine to another machine only transfers the application instance. This provides a lightweight solution in comparison with the VM + web applications approach. As shown in Figure 1(b), this approach can be built in either non-virtualised (e.g. traditional IDC) or virtualized (e.g. Amazon EC2) environment, in the other word, it can be built on either in PMs or VMs. On top of the resources, there is a cluster of application servers that are for hosting web applications of cloud users.