Software defined networking and many of the accompanying trends have revolutionized the network. In Part 1 of this series, we discussed how white box switches are driving the network to increased commoditization. The flexibility and programmability provided by white box switches and other programmable hardware crucial to SDN opens the market up to new competition, forcing innovation, lower costs and more flexibility. In addition to more flexible hardware, there is more flexibility in software, allowing for services to be more closely integrated with the system.
These services, which could range from firewalls and load balancing to deep packet inspection and quality of service, historically have been completely separate from the switch and been quite costly to add to the network. The high degree of programmability and flexibility that white box switches provide allows these value-added services to be quickly, easily and most importantly, cheaply programmed onto the switch itself. Network operators can now completely change the character of their network by integrating new software and services into the switch to meet new and changing needs.
In order for this new class of network service switches to reach their full potential they will require higher performance processors included in the switch and data plane connectivity between the switch and the processor commensurate with the targeted level of network services performance. This concept should be familiar to network data center operators. Server computers come in all shapes and sizes depending on the workloads they are optimized for. Some servers provide significant I/O connectivity, others large amounts of memory, processing or algorithm acceleration. We should expect that white box switches would also be offered with different amounts of networking connectivity and processing performance.
Most classical network switches include an embedded communications processor that is coupled to the switch ASIC via PCI Express. This connectivity is used to configure the switch ASIC and to support exception traffic from the switch to the processor. When the switch encounters a network packet that it does not know how to process, it will send it to the processor via the PCI Express link and the processor will determine the correct course of action for the packet and them may reconfigure the switch so that subsequent packets will be handled by the switch itself. For classical switching systems operating on the hundreds of gigabits a second to terabits per second, the multicore communications processor will typically have 2 to 4 gigahertz class cores performing this work.
For network services switches much higher performance processors will be required and datapath connections will need to be increased. For example to support a 20 gigabit firewall or IPS service, the multicore communications processor will need to have the capacity to perform this level of processing (in addition to the classical switching processing). It will also need to have at least 20 gigabits of dedicated data connectivity between the switch ASIC and the multicore communications processor. This can easily be achieved by coupling 2 or 4 10Gigabit Ethernet ports between the switch ASIC and the multicore communications processor.
By coupling the switch ASIC and the multicore processor together in this way, a very flexible platform is created that can fully support classical switching applications, fully support SDN requirements and also provide for adding services that are traditionally contained in separate appliances. Applications such as firewalls, application identification, load balancing and IPSEC can be implemented as services in this network service white box switch.
This network service switch concept opens up the networking market to an entirely new level of competition and capabilities. We should expect many different kinds of white box switches, all running a common software environment based on Linux with the network services processing occurring through a data plane API such as DPDK or Open Data Plane (ODP), running on hardware platforms of varying processing capabilities and cost points. Customers will have greater choices and will no longer confined to a single vendor’s ecosystem, picking and choosing their hardware, software and services to meet their unique network needs, whether those are today’s needs or needs that will appear in a few years. These new levels of capabilities and competition will eventually bring about not only lower prices but also increased innovation.
There are still a lot of questions that remain – what about standards? What role will Open Flow play? How will support be handled? What interoperability issues might arise and how will they be dealt with? What will the long-term impact be on the networking market size? But it is clear that this trend is real and has already begun to move the industry in a positive and exciting direction.