Telecom infrastructure is one area where X86 architecture hasn’t dominated historically. Infrastructure gear is spread across MIPS, Power, and SPARC architectures, with some X86, and a relative newcomer: ARM, already claiming 15% share. That’s a stunning figure considering only a bit less than 5 years ago, there wasn’t even a viable 64-bit ARM server chip – ARMv8-A officially arrived on the scene in October 2011. We’re talking about 20 year lifecycles and a boatload of software involved in switching architectures.
What has happened in the last 21 months is even more stunning. As Bob Monkman, enterprise segment marketing for ARM, said in a recent Light Reading webinar, 85% of telecom infrastructure chip shipments come from current ARM partners shipping either legacy architectures (like Broadcom, Cavium, and NXP) or new ARM-based parts (extend that list with AMD, Applied Micro, Marvell, HiSilicon, Mellanox, Qualcomm, and Xilinx among others).
Why the relatively sudden shift? 21 months ago, the OPNFV initiative within The Linux Foundation launched. Folks had been trying to evangelize open-source software in telecom for years, dating back to the Service Availability Forum, the SCOPE Alliance, and others. Unseating a century worth of proprietary efforts in telecom is a huge undertaking, and there are a lot of moving parts from both the legacy and new innovation to deal with.
OPNFV looks like it has broken through, finally opening the door for collaborative development rather than just marketing lobs of open-source code that vendors have to port and integrate. OPNFV of course realizes NFV constructs, but more importantly they integrate components from upstream infrastructure projects including OpenDaylight, Ceph Storage, OpenStack, KVM, Open vSwitch, and others. They are also harmonizing the split between DPDK (Data Plane Development Kit) that originated on X86, and ODP (OpenDataPlane) which came out of the ARM-centric Linaro efforts.
And, there is the end-to-end angle for telecom. 95%+ of mobile devices are ARM-based, so creating a service quickly that talks to those devices could depend on compatible software IP. Monkman said opening this event that the carriers are trying to deploy services in the same day instead of the industry-average 8 months of application development time. Architecturally, NFV is a huge part of that, but from an ecosystem perspective, ARM and its partners can tool up and get telecom vendors to the goal line much faster.
There is also synergy with other Linux efforts; we’ve already mentioned Linaro. Another trend in the Linux community is a move toward containers and microservices and away from heavy VMs and managed services. Lighter-weight network edge virtualization and the “micro-cloud” or “fog” is a growing trend not only for telecom services, but for IoT infrastructure. This will accelerate with the debut of 5G technology, regardless of the specifics of the finalized standard. One key point Enea’s Joe Kidder makes in this webinar is the same stack runs on the 4 core Broadcom NorthStar 2, the 8 core NXP LS2085, and the 48 core Cavium ThunderX, meaning code can be deployed across that spectrum quickly.
Another example Kidder offers is OpenFastPath, a new open-source user-space TCP/IP stack. What’s the big deal? Most TCP/IP stacks are kernel space, tightly coupled with a processor architecture (often locked to a particular core) and subject to the whims of Linux kernel scheduling and loads. OFP does something rather cool: it scales with the number of cores to 10Gb Ethernet line rates.
That’s just a sample of what Monkman and Kidder share in this webinar, not an NFV technical primer but a solid overview of the status of OPNFV on ARM:
What comes next is the “Colorado” release of OPNFV, headed from the experimental labs to field trials just in time for 5G. Enea is plowing ahead on security, hardening, performance, edge use cases, and interoperability testing – all on ARM architecture. ARM’s stated goal is “VNF ubiquity”, meaning having all the use cases covered so operators can assess the value propositions of ARM architecture running all this open-source code, end-to-end.
Granted, this does not impinge directly on Intel’s high-volume server space, but the lifecycles and value-add in telecom infrastructure means higher margins and steady revenue once design wins are secured. ARM not only has a head start here, but I’d submit they are likely to run everyone else over within the next three years and become the dominant player in telecom infrastructure – of course, with the help of their massive and deeply experienced ecosystem.