The usage of broadband access networks has skyrocketed over the last decade, which has led to the emergence of novel web applications. This scenario has also heralded the birth of revolutionary Internet usage via free-of-charge applications through datacenters. In most cases, the end user enjoys the transparency of these datacenters. Therefore, the trend is shifting from traditional carriers to content-based carriers like Google and Comcast, which are non-traditional carriers.
Private bandwidth usage escalated from 2002 to 2008 to stand at approximately 47% annually. The network landscape is reshaping fast with the entry of content distribution networks (CDN) and cloud computing and providers of such contents like Google, Microsoft, and Amazon among others have become key players in optical network deployments.
Internet-style computing underscores the connectivity of different mega datacenters located in disparate regional locations using large-capacity networks, thus forming a cloud, hence the term âcloud computingâ. To users accessing the cloud of networks, the datacenter network stands out as a solitary computer, which can be accessed via oneâs local network providers through Internet Points of Presence (POPs). This form of computing comes with numerous advantages including scalability, automatic backup and easy data sharing platforms. In addition, users incur low costs in terms of operational expenses and the capital needed for computing machinery.
Intra-datacenter communication is a web of interconnected datacenters in a bid to realize optimal performance coupled with reduced costs. However, in a bid to understand intra-datacenter communications, it is important to define a âdatacenterâ, which is a âmassively parallel super-computing infrastructure, which consists of clusters with thousands of servers networked togetherâ (Lam et al. 33). In intra-datacenter communications, different servers are grouped into racks, with each rack carrying a maximum of 40 machines. Servers hosting machines in the same rack share connectivity via a top of rack (TOR) switch.
The switches hosting different racks are then connected using cluster switches, which ultimately connect different racks coupled with forming clusters at the same time, ultimately making warehouse scale computing. Conventionally, the best prototype for intra-datacenter network would be the one that connects all servers within a datacenter, but such a design is very costly. Therefore, the commonly used intra-datacenter network entails cluster interconnections of switch fabrics with different racks.
The density of the ports in a switch determines the number of servers that can be linked to a certain switch. In addition, âthe scaling of the centralized controller will limit the number of nodes that can be managed in one clusterâŠmaximizing the switch port density and the number of nodes of a cluster is an important concern for intra-datacenter networking equipment designâ (Lam et al. 34).
Conventionally, datacenters are distributed across different geographical locations by considering disparate factors like power stations and costs. For instance, rural areas with reliable power supply and low-cost land are the best-suited geographical areas for datacenters. However, the dispersed locations of datacenters pose a challenge whilst moving traffic to population centers from datacenters as long-haul optical networks are needed due to lack of fibers in inter-datacenter communication, which underscores communication between datacenters. This form of communication involves huge volumes of data, which calls for efficiency in terms of capacity and spectral aspects. In addition, inter-datacenter communications are costly due to the nature of work and volumes of data involved.
Therefore, in cases where fibers are readily available, it is advisable to utilize them instead of expensive technology to achieve the same purpose. However, in areas where fibers are not available, adopting modern transmission technology is inevitable. Nevertheless, whichever method is used, network operators seek to minimize the cost per bit while maximizing the total capacity. Therefore, commercial equipment vendors should go beyond C-band, while infrastructure providers should âdeploy low-loss, low-nonlinearity, and large-core optical fibers in new fiber buildsâ (Lam 39).
Works Cited
Lam, Cedric, Hong Liu, Bikash Koley, Xiaoxue Zhao, Valey Kamalov, and Vijay Gill. âFiber Optic Communication Technologies: Whatâs Needed for Datacenter Network Operations.â IEEE Communications Magazine 2010: 32-39. Print.