Development of IT technology has always been a multi-track process. Servers footprints shrink while gaining more sockets and bigger and better data ports, while storage systems acquire fancier controllers and more sophisticated array designs. But the real action is almost always on the microscopic level: silicon processors in the case of servers and desktops, flash, RAM, and drive platter technology for storage.
The same holds true for networking. High-speed standards and protocols can improve data rates. So can advanced switching and routing architectures. But each generation of network processing technology not only boosts speed, but also enables more streamlined networking infrastructure capable of handling bigger loads – the classic “do more with less” paradigm.
And lately, we've witnessed a flood of network processor developments. Three, to be exact, but this is such a specialized field that three constitutes a flood.
The first came out about a month ago with Huawei’s introduction of the S12700 Agile switch. The device enables broad SDN architectures across campus networks, animpressive achievement in itself. Under the hood, however, things get even more interesting. The S12700 switch uses Huawei's own microchip, the Ethernet Network Processor (ENP), which boasts a streamlined programming architecture that allows Huawei to field customized network services in as little as six months. This adds a new level of physical-layer flexibility to meet rapidly changing data usage patterns. It also enables the enterprise to more easily integrate diverse networking elements, such as mobile devices, into unified network architectures.
Meanwhile, Intel recently introduced the Xeon E5-2600 v2, which boasts up to 12 cores on a 22 nm process. Although not a networking processor per se, it does allow the enterprise to create high-volume servers that can consolidate network applications and take on tasks that currently run on costly offload engines and acceleration systems. The device supports Intel’s Open Network Platform (ONP) reference design and the Open vSwitch platform to consolidate disparate network architectures into a common fabric.
Traditional networking platforms are geared toward connecting data users either with other users or stored volumes. Increasingly, however, enterprises must deal with machine-to-machine traffic, the scarcely noticed, unstructured data fueling the Big Data revolution. Cisco hopes to tap this market with the new nPower processor, a 400 Gbps device that the company calls the key to the “Internet of Everything.” The first model in the line, the X1, holds more than 4 billion transistors and 336 cores. Though Cisco ostensibly intends it for large carrier networks, it won’t be long before large enterprises or cloud providers find a use for a massively parallel core switch, particularly as SDN and other developments stress network agility over raw throughput.
This kind of network processing power is impressive, but Light Reading's Sterling Perrin offers a word of caution as we broach the 400 Gbps level: network performance on that scale is very difficult to benchmark. According to the most recent lab tests, packet loss tends to rise with ultra-fast throughput, so you may be able to access a distant file that much quicker, just not all of it. And the problem becomes more acute when dealing with the small-packet data that characterizes web transactions and machine-to-machine communications as processors become overloaded with table queries and other tasks.Still, it’s fair to say that just about every new processor generation has faced its share of engineering challenges on the way from the lab to full production environments, so there’s no reason to believe that network devices have hit the wall.
It is also true that the only certainty in networking these days is that change is, and will remain, constant. Traditional networking silicon served admirably throughout the client-server era and well into the virtual age, but with SDN, mobility, unified architectures no longer on the horizon but right here in front of us, it’s time to revamp network architectures right down to the very core.