gains in terms of performance scalability and agility to their users. By leveraging cloud computing, an organization can rapidly deploy applications where the underlying technological components can expand and contract with the natural ebb and flow of the involved business lifecycle. Traditionally, once an application was deployed, it was bound to a particular infrastructure until the infrastructure was upgraded/improved. The result was low efficiency, poor utilization, and limited flexibility. Cloud enablers such as virtualization and grid computing allow applications to be dynamically deployed onto the most suitable infrastructure at runtime. Cloud computing takes these concepts further, by allowing more automated resource and workload management practices. This elastic aspect of cloud computing allows applications to scale and grow without needing traditional “fork-lift” upgrades. Like any new paradigm, Cloud computing represents an architectural shift from the traditional distributed computing approaches. Such a shift is best described by the addition of a new and as transparent as possible middleware layer on top of the existing computer and device operating systems that we can call a Cloud Operating System (COS). It can be considered as a network operating system running atop a cloud, i.e. a hyper network of computers. As its name suggests, this kind of runtime platform lets users write applications that run in the cloud, or to use services provided from the cloud, or both. But the transformation that cloud computing makes possible goes beyond simply running applications on a virtualized platform built on someone else’s hardware. It extends the computing model with the transparent utilization of a platform that the provider has created, and which, to some degree, abstracts the essence of scalability and distributed processing. More generally, the concept of cloud computing can incorporate various computer technologies including web infrastructure, Web 2.0, and many other emerging technologies. People may have different perspectives from different views. For example, from the view of the end-user, the cloud-computing service moves the application software and operation system from desktops to the cloud side, which enables users to plug-in anytime from anywhere and utilize large-scale storage and computing resources. On the other hand, the cloud-computing service provider may focus on how to distribute and schedule the computer resources. Nevertheless, storage and computing on massive data are the key issues for a cloud infrastructure.

8.3  Improved Network Support for Cloud Computing

The promise of cloud computing is ubiquitous access to a broad set of applications and services, which are delivered over the network to multiple customers. Such services are essentially offered through interfaces available within the “clouds,” rather than spread over the single computers connected through the Internet. On the other hand, such cloud infrastructures, because of their high degree of abstraction, have the potential to introduce unpredictable performance behaviours. In fact, while sharing the resources available on a large distributed infrastructure can average out the variability of individual workloads, it is extremely difficult to predict the exact

performance characteristics of your application at any particular time. Like in any shared infrastructure, varying individual workloads, resource demands, of and network load conditions, can result in unpredictable performance behaviour of the combined applications. Furthermore, as cloud computing enables users and applications to store all their data on the network, handling and moving large volumes of data within the cloud or between the users and the cloud may become a challenging issue. Consequently, cloud-service providers must guarantee that data are processed automatically and transferred transparently when and where they are needed. Also important to the notion of cloud is the automation of these tasks. An environment in which the system requires human intervention to allocate bandwidth on communication links or resources to processes is not a cloud: it is simply a data center. Integrating an accurate network view into the cloud management in order to support these types of services would make the cloud more flexible and also increase the efficient use of the available resources and communication infrastructure. The underlying network architecture building the foundation for cloud computing consists of interconnected server farms within the data centers and a high-speed transport network providing connectivity to remote and backup sites. These high-speed connections form the backbone of the cloud network and are required to run at highest bandwidth with lowest transmission latency, and in general, according to a properly defined QoS degree. Cloud-computing resources can be made accessible through the public Internet, private high-performance networks, and often through a hybrid mixing of the two. Providers and users of cloud services must understand the performance, redundancy, and cost associated with all the available options, because not all the applications that have to be run on the cloud have the same features: some will only require the basic capabilities available on the public Internet or traditional public connection services, while others may require the enforcing of specific network performance constraints.

8.3.1  Why the Internet is Not Enough?

The public Internet is the simplest choice for delivering cloud-based services. In this model, the cloud provider simply purchases Internet connectivity and its customers access the services via their own Internet connections. However, modern highperformance applications are raising communication and bounded-time execution requirements that the public Internet cannot meet neither at the present nor even in the foreseeable future. In fact, the traditional Internet-shared network paradigm is based on the best-effort packet-forwarding service that is a proven efficient technology for transporting burst transmission of short data packets, e.g., for remote login, consumer-oriented email, and web applications. Unfortunately, this is not enough to meet the challenge of the large-scale data transfer and connectivity requirement of the modern cloud-based collaborations. More precisely, the traditional packet-forwarding paradigm, based on statistical multiplexing, is not scalable in its ability to rapidly move very large data quantities. Making forwarding decisions every 1,500 bytes is

sufficient for emails or 10–100 KB web pages. This is not the optimal mechanism if we have to cope with data size of six to nine orders larger in magnitude. For example, copying 1.5 TB of data using packet switching requires making the same forwarding decision about one billion times, over many routers along the path. Internet-based cloud infrastructures lack the scalability features required by modern data-transfer- intensive applications in several aspects: network resources cannot be reserved in advance, bandwidth is too low, QoS is not guaranteed, and hence, neither application success nor a bounded completion time can be ensured to the users. Particularly, reservation of connectivity resources is needed to facilitate the transportation of enormous datasets between distant sites within the cloud in predictable times. Clearly, this cannot be easily achieved in traditional packet-switched networks, where the resource needs are usually not known and hence cannot be planned in advance.

8.3.2  Transparent Optical Networks for Cloud Applications: The Dedicated Bandwidth Paradigm

Creating a dedicated circuit over several available high-speed links will be a much more effective multiplexing technique for large data transfers. Consequently, there is the need to develop new architectures and services that support cloud infrastructures in association with emerging networks technologies, having the potential to always provide in advance the available large bandwidth pipes with a capacity of several orders of magnitude beyond that of today’s communication infrastructures. Only a modern optical transport network provides the capacity and bandwidth needed to support these demanding cloud-computing applications. Accordingly, in order to achieve connectivity resources in terms of bandwidth and quality of service (QoS) when and where the applications need, it is necessary to perform advanced provisioning of end-to-end dedicated optical “virtual circuits” through the network, implemented on properly reserved wavelengths on the available Wavelength Division Multiplexing (WDM)-based optical transport infrastructures. In detail, according to the WDM paradigm, the optical transmission spectrum can be carved up into a number of non-overlapping wavelength bands, each supporting a single communication channel operating at whatever protocol or rate one desires (protocol and bit-rate transparency). Thus, by allowing multiple independent channels to coexist on a single fiber, we can make the most of the available optical infrastructures, with the corresponding challenges being the design and development of appropriate network architectures, control-plane protocols, and algorithms that make these connectivity resources available to cloud applications. As such network control plane provides an advance reservation capability to the circuit, the data-intensive applications can be guaranteed to achieve certain bandwidth and QoS in specific time slots. This can be considered the most promising mechanism to meet all the future data transfer demands from applications running on the cloud through the provisioning of huge amounts of cheap bandwidth through dedicated end-to-end connections fulfilling the proper QoS requirements.