Specialized Network Devices Will Not Be Going Away (Anytime Soon)
December 4, 2014 Leave a comment
- Not all parts of the service provider network infrastructure can benefit from virtualization; analysis of the IP and optical core and the aggregation layer, for example, points toward the continued use of specialized hardware/silicon and networking software.
- The operational model service providers are looking to is more about service agility than it is about running everything on an x86 blade; dedicated hardware equipped with high-performance network processors continues to be the ideal choice to support high-speed/high-density 100G Ethernet and optical super-channels.
As operators seek to add programmability and agility into their services (and, thus, their networks), it is clear that the discussion regarding which functions should be virtualized versus those that should continue to be performed by specialized hardware/silicon solutions will continue, even as vendors such as Intel and Cisco push the envelope to ratchet up the performance of their x86-based platforms with multi-core architectures and new high-performance development environments. Virtualization certainly makes sense for anything that is heavily weighted toward compute and storage and for data plane applications that require moderate throughput. However, in the IP core network, where the demand is to support multiple terabits of throughput, custom solutions comprised of vendor-specific silicon and hardware will continue to provide a viable, if not the only, solution.
The same can be said for the network aggregation layer, where the task is to collect multiple traffic streams, bundle them into a few high-speed links and forward them to the service edge for processing. Alcatel-Lucent’s recent 7705 SR-a announcement is targeted at addressing this need, and it can also provide basic services.
For the access layer, however, we are seeing a bit of a shift from the traditional backhaul model, where traffic is sent on to the service layer or mobile core for processing, to one with localized processing (via an x86 blade of sorts); this architecture is seen in solutions from Nokia (with its Liquid Apps) for the mobile network and RAD (with its distributed NFV) for fixed line services. In all cases, the need for network programmability is growing to support service agility and to conquer dreaded increases in complexity.