Wind River has introduced what it calls an "ultra-fast" packet-forwarding platform for overburdened telecom providers.
Just how fast is ultra-fast? According to a Wind River release, when running on Linux on an Intel Xeon 5500-based reference board, the new Network Acceleration Platform managed iPv4 forwarding at a rate of 21 million packets per second - and that's using just four threads.
Wind River, which was acquired by Intel last June, claims that this level of performance is "more than five times" that of the same hardware running native Linux in symmetric multiprocessing (SMP) mode.
The operating system used in the control plane of Wind River's network acceleration engine is either Wind River Linux or VxWorks. The platform's data plane runs on the lightweight Wind River Executive OS. The combination of that small-footprint OS and the company's multicore-optimized networking protocol software that enables iPv4/iPv6 packet forwarding, a Wind River spokeswoman told The Reg, adds up to the platform's packet-acceleration engine.
Wind River product manager Felix Baum told The Reg that the platform is intended to be flexible such that customers writing for it - telecom customers, that is - can choose to write their software to use the cores at varying levels of multithreading.
"In some cases," he said, "it is much faster if you have a dedicated algorithm that runs a single thread - you receive a packet, you process the packet, and you push it out without getting interruptions, without getting sidetracked by other things."
However, doing so may waste a multicore processor's horsepower: "In many instances, these processors - these many individual cores - are way too powerful, and they can push packets and still have some slack." That slack can be used for any number of things - including using additional threads for more network acceleration.
The Network Acceleration Platform can act as an arbiter between the CPU core and the networking silicon. "In a perfect world," Baum said, "what we can do with our operating system and solution here is that we can actually go and set up the silicon so that we dictate the rules - we tell it what to do." And what it can do might be deep-packet inspection, virus-checking, and the like.
"For example, we can specify and tell you - the silicon - that when you receive a packet, before you give it to me - the core - to process it, to look at the third bit and depending on it, do this, and look at the fifth bit, and do that. And here's the pattern I want you to match, and if any part of the packet matches this - a virus, for example - then go and do these things before even handing it off to me, the core."
And the Network Acceleration Platform manages the entire system. "We give users the tools to actually configure the whole system so it all happens once they flick the switch and the system boots. They set up these devices in a particular fashion to help customers do what they're trying to do."
And what those customers are trying to do is dig telecoms out of the hole they've dug for themselves by flooding their infrastructure with data traffic - an infrastructure that was originally designed for voice traffic.
In a Tuesday blog post, Wind River networking product line manager Mark Guinther noted the clear need for the type of performance offered by multicore telecom setups. "Those are the times we grumble about our network service, or device, or both. For service providers, unhappy customers lead to increased churn and less revenue."
Guinther's solution plays into the strengths of his Intel masters. "Networking equipment designed on multi-core processors offers the increased performance capability, lower space, weight and power requirements, and scalable designs that service providers need to win customers and reduce churn."
Better that any churn should not be among customers, but rather in the data and instructions being passed about by Wind River's Network Acceleration Platform, third-party communications silicon, and multicore processors from the Wind River's parent, Intel. ®