Challenges in Future Intra-Data-Center Networks and Possible Solutions

Challenges in Future Intra-Data-Center Networks and Possible Solutions

Muhammad Ishaq (Pakistan Institute of Research and Development, Pakistan), Mohammad Kaleem (COMSATS University – Islamabad, Pakistan) and Numan Kifayat (KAIST, South Korea)
DOI: 10.4018/978-1-5225-9767-4.ch005


This chapter briefly introduces the data center network and reviews the challenges for future intra-data-center networks in terms of scalability, cost effectiveness, power efficiency, upgrade cost, and bandwidth utilization. Current data center network architecture is discussed in detail and the drawbacks are pointed out in terms of the above-mentioned parameters. A detailed background is provided that how the technology moved from opaque to transparent optical networks. Additionally, it includes different data center network architectures proposed so far by different researchers/team/companies in order to address the current problems and meet the demands of future intra-data-center networks.
Chapter Preview


The facility used to house massive amounts of computing, storage and network resources like servers, hard drives and bandwidth is called a Data Center (DC). These resources can either be used by the DC operators to deploy their own services or be rented to their customers. These customers are usually Small and Medium Enterprises (SMEs) which have reduced resource requirements and would find very expensive to deploy and maintain their own infrastructure. Therefore, the DC operators provide Infrastructure as a Service (IaaS) (Buyya et al., 2009) to their customers.

The main issues which DCs address are linked to the fulfillment of the Service Level Agreement (SLA) (Bouillet, Mitra, & Ramakrishnan, 2002) which their operators sign with the customers and the maximization of the profits they obtain. In such a context, the key challenges DCs operators have to face are:

  • Scalability: The capability of being able to increase the number of housed resources and bandwidth.

  • Fault Tolerance: The capability of being able to withstand failures without producing an impact on the service.

  • Cost Effectiveness: The capability of reducing the amount of required resources. This can be achieved through the use of several virtualization technologies, abstracting the physical resources into several virtual (or logical) resources. This allows the DC operators to optimize the resource usage by providing to each customer only what he needs and pays for.

  • Power Efficiency: The capability of minimizing the power consumption produced by the resources.

By taking a closer look into intra Data Center Networks (DCNs) it can be observed that they intend to follow these same principles. An intra DCN’s function is to allow resilient, high bit-rate and low-latency communications between the DC’s computing and storage resources.

This is a critical task since a failure (or congestion) in the network would degrade the performance of the connections or directly block them. If this happens, the outcome (from the user’s point of view) is the same as if the computing or storage resources were down since his request cannot be attended. An example of this would be the communication between a virtual machine and a storage server (Figure 1).

Several forecasts, among them the Cisco GCI 2016 (Figure 2), indicate that the global

DC traffic will grow from about 6.8 Zettabytes/Year on the year 2016 up to approximately 20.6 Zettabytes/Year on the year 2021 experiencing a threefold increase. Moreover, the expected Compound Annual Growth Rate (CAGR) for global DC traffic is around 25%.

Furthermore, in such a period of time, the 73% of the global DC traffic is expected to be intra DC traffic (Figure 3). This forthcoming substantial growth in intra DC traffic will push to the limit the scalability and performance of currently deployed intra DCNs’ architecture.

Figure 1.

Virtual machine to storage successful and failed communications

Source: Predieri et al. (2013)
Figure 2.

Global data center IP traffic growth

Source: Cisco GCI (2016)
Figure 3.

Global data center traffic by destination

Source: Cisco GCI (2019)

Key Terms in this Chapter

Micro-Electro-Mechanical Systems Switches (MEMS Switches): MEMS optical switches are mechanical devices that are used to physically rotate the mirror arrays and redirecting the laser beam to make a connection between the input and the output. As they rely on mechanical systems, so the reconfiguration time is a few milliseconds. Currently, the commercial available MEMS optical switches can support up to 32 input/output ports.

Splitter and Combiner: Splitter is a passive optical device that is used to split the optical signal from one fiber to two or more fibers. On the other hand, combiner is utilized to combine optical signals from two or more fibers into a single fiber.

Arrayed-Waveguide Grating (AWG): AWGs are passive optical devices that are not dependent on data-rate and are used to route each wavelength of an input to a different output (wavelength w of input i is routed to output [(i + w - 2) mod N]+1 , 1 =i= N , 1 = w = W , where N represents the number of ports and W is the total number of wavelengths). AWGs are used as de-multiplexers to separate the individual wavelengths or as multiplexers to combine them in WDM communication systems.

Coupler: Coupler is a passive optical device that is utilized to aggregate and distribute signals in an optical network, but it can support multiple inputs and outputs. For example, a 2x2 coupler gets a fraction of the power from the first input and gives it on output 1 and the remaining fraction on output 2 (similar goes for the second input).

Complete Chapter List

Search this Book: