In the third and last part of a three-part series, we conclude the WAN Summit London 2015 discussion about SDN and NFV with a look into the "white box revolution," an idea that promises a vendor-agnostic solution in which open source software operates on generic server and switch hardware.
The speakers on the panel were:
Chairing the panel was Erik Kreifeldt from TeleGeography.
Ray Watson from Masergy explains it as follows: “one of the biggest promises of SDN and NFV is what’s been referred to the ‘white box revolution’ which is that you can go build a server yourself, put a custom Linux kernel on there of some type of hypervisor-capable software and immediately be able to virtualize these functions from all these different vendors and different services that are out there today.”
The panel agreed that the picture is much more complex than that; simply taking what you have in hardware and then sticking it in a virtual machine is not going to deliver the throughput and can’t really be considered true virtualization.
Some vendors might be trying this approach in the short-term in order to bring a virtualized product to market, but the panel were much more interested in approaches to building stuff from the ground upwards as a virtual form factor and building it to run just on commodity hardware.
Malcom Orr from Colt argued that “When we look at the optical space, when we look at the layer 1 services, there’s no way we’re going to run those on a hypervisor. We might run them on commodity platforms once we can get the physical mix sorted out but we just won’t get the throughput on virtualised form factors and there’s no need to virtualise them.”
Ray also warned of the dangers of conflating the two issues; just because you’re being offered a VNF doesn’t mean it is–or can–run on commodity hardware: “The scary thing is there is all these vendors which start off by trying to offer you an eight-core Xeon box with a terabyte of RAM and 2 solid state drives, right? I mean, it completely obliterates the entire model of power consumption and cost savings that are out there. […] Customers really want to see some economic benefit to doing it.”
Ray said: “The leader in this space and one of the most vital pieces for these types of deployments is to be based on Intel’s DPDK. Their Data Plane Development Kit is a really big game changer when it comes to performance and optimization.”
“We were shocked to find that many of the vendors that we shopped for VNFs from sent us images that weren’t even optimized for DPDK,” continued Ray. “They literally ported their ASIC-based proprietary stuff to an image that would ‘kinda work’, right? And we weren’t getting the performance there.”
Tom Homer of Telstra pointed out that latency is a really big issue in some industries, especially financial services, so there will be some things that can’t be virtualized, at least in the short term.
Malcolm suggested that, just as most people developing for the cloud now know that the infrastructure is not resilient so they build the required resilience into the application, building application resilience will be key to the white box revolution in the network space. “Once that application resilience is there you can actually run it on something you built yourself,” he said.
Ray agreed that while we aren’t quite at the point where anyone in a garage can build their own production-level switch or router yet, it is coming – and Intel and OpenStack and OpenFlow are leading the way.
“Just like five years ago I would have told you ‘don’t buy a router that can’t do IPV6’, I would encourage those of you who are thinking of doing something in the white box space within your data centre or in your WAN to make sure that it is capable of working with OpenStack and OpenFlow – that’s the baseline for all that you want to do in that space.”
The panel’s key takeaways on the white box revolution are:
Find out how the first part of the debate went here: Exploring the Potential of SDN and NFV and Virtual Reality: NFV Choices.