Issue link: https://cp.revolio.com/i/141720
As data center networks continue to evolve from being connectivity oriented to becoming more services oriented, there will be a paradigm shift in the way network fabrics are architected and built. Instead of building layered decision-making complexities into switches and routers, the new paradigm will require minimizing complexity in network elements by increasing homogeneity, as well as simplifying configuration and programming capabilities. A major driving force for this new paradigm is the change occurring in traffic patterns. As traditional dedicated server farms make the transition to becoming more virtualized, and as network-attached storage (NAS) and storage-area networks (SANs) continue to converge, traffic patterns are changing in both the north-south and east-west directions. Network security is yet another aspect that is undergoing a major transformation from being periphery-based to becoming more pervasive and dynamically deployable throughout the data center. The underlying driving force for these changes is, of course, the data deluge and the resulting need to scale data centers more cost effectively. The best way to do that is to simplify services orchestration in a more commoditized operational model, and this is driving the desire for the greater flexibility afforded by software-defined networking (SDN). Whether using SDN or some other approach, the major challenge in re-architecting next-generation data center networks will involve making them more services oriented. But before exploring how a services-oriented networking (SON) architecture will enhance scalability, it is helpful to understand some key aspects of how data center networks scale today. Scaling Data Center Networks Today The traditional means for scaling data center networks is to create new subnets by adding domains, as Figure 1 shows. Each individual domain is responsible for forwarding and otherwise managing traffic for its members. A second network tier is then used for managing all of the inter-domain traffic. A third tier is introduced with the need for an enterprise network, as well as for connectivity to the Internet and cloud-based services. This third tier further increases the overall network's complexity owing to the need to accommodate various enterprise, carrier, service-provider and data center network requirements. Such a connectivity-oriented hierarchical data center network works well for relatively static workloads where the compute and storage infrastructures are purpose built for specific applications. The network's data plane is based on the International Standard Organization's proven seven-layer reference model, with switching performed at Layer 2, routing at Layer 3 and some semblance of services performed (more or less) at Layers 4 through 7. The traditional tiered data center network architecture works less well (and often quite poorly) as workloads become more dynamic. The main problem occurs in the control and management planes, which require each network element to operate in a mostly autonomous fashion. The result is a need for a plethora of protocols with some complex management interactions, especially for network state discovery and dissemination. Figure 1: Existing data center networks are connectivity oriented, requiring them to be scaled in hierarchical tiers. Cloud Tier 1 Scale Scale Tier 1 Tier 2 Internet Tier 3 Tier 1 www.datacenterjournal.com THE DATA CENTER JOURNAL | 21