Intel has ambitions to turn modems into virtual servers and reinvent broadband A few cores in your modem and OpenStack driving network function virtualisation
====================== [The Register has been very negative and critical of Intel over the last several years. So I was surprised to see something favorable to Intel from them...] ======================
Intel has assembled a stack of technologies it thinks can give broadband modems a brain implant and change the nature of home broadband services.
At the core of Chipzilla's ambition, as Intel folk have explained to The Reg, is a plan to put x86 chippery into customer premises equipment (CPE) – the modem/router homes and smallish businesses use to connect to the internet.
Such devices are currently pretty dumb, pretty cheap, and pretty much MIPS-powered. Plenty run embedded Linux and a few have remote control apps, but the Linux is locked away and there's minimal scope for customisation at an individual level. Only the brave-hearted ever try re-flashing their modems with open firmware like DD-WRT.
Intel thinks that building x86s into CPE devices gives them the grunt to become more interesting and looks to be taking steps to make this happen more often.
The company already bakes not-too-shabby Atom cores into its PUMA range of DOCSIS 3.0 cable modems. Work is under way on DOCSIS 3.1 kit, too, to help deliver gigabit cable Internet performance. Intel's also acquired Lantiq, a big player in DSL modem system-on-chips. Lantiq's already playing in G.fast, the gigabit-speed successor to VDSL. It's hard to imagine Intel have bought Lantiq without pondering pairing its SoCs with x86 chips.
Faster and more powerful CPE devices are a bit of a so-what proposition. Until one considers that Intel packs full x86 cores into PUMA kit and will probably do the same for Lantiq. And by full cores we mean CPUs with all of Intel's virtualisation-friendly and networking bits included.
Virtualisation is important to this and Intel's story because it gives carriers a new way to deliver services to subscribers.
For years, internet service providers (ISPs) have been purveyors of bit pipes and little else. Plenty have been able to charge a buck or three a month for services like extra security, hosting or storage, but over-the-top services have nearly always won the day, as shown by the fact that Dropbox rules and ISPs like Australia's iiNet have exited the cloud storage and security business.
CPEs that can run virtual machines give the carrier a platform to push services onto the kit, if they can't be run in the cloud. Lightweight hypervisors aren't hard to find and won't strain a multi-core Atom box unduly.
Consider WAN optimisation, for example. Purveyors of that technology nearly always need to deploy software on-premises and in the data centre. Getting your average family or small business to install an app on every device to make that happen is asking for trouble.
Running it in an on-premises virtual machine spawned into the CPE with the only customer intervention being to click a box saying “Do you want faster downloads for $1/month?” is a lot more likely to have a happy ending.
Or consider firewall-as-a-service. Today, an ISP offering extra security probably routes customer traffic through an actual firewall. A smart CPE could run that firewall on premises. Or an ISP could deploy network function virtualisation (NFV) and run the firewall as a VM somewhere in its own infrastructure, removing the need for a dedicated firewall. The smart CPE gets to run a VPN to the firewall, to secure the whole connection.
Either way, it means the customer gets better security, without having to learn firewall administration.
How to get a firewall, WANop, or any other VM into a CPE? Enter OpenStack, which has no trouble spawning and managing VMs all over the place.
OpenStack as VMs-everywhere wrangler
Intel's already working on ways to get OpenStack running at carrier scale. The company last year teamed with HP to create a carrier-grade version of the Helion cloud aimed at making it possible to deliver virtualised services to CPEs. HP uses Wind River's Linux in this kind of gig. And Wind River is owned by ... Intel, which has also released the Open Network Platform, a white box switch and network function virtualisation play. Oracle works with Intel on that platform.
Intel also sells Xeons to power OpenStack clouds, plus silicon to power those Open Network white box switches carriers can use to create their on-premises OpenStack rigs if they choose that route.
OpenStack's critical because it is free to acquire and scales well enough to contemplate it as the platform for deployment of many, many VMs to CPEs. Carriers might be able to do large-scale NFV and deployment of VMs to CPEs with VMware, but almost certainly at a far higher cost than with OpenStack.
This all starts to stack up (pardon the pun) interestingly if one considers the internet of things, because something has to collate and groom all the data that smart thermostats and mould-aware fridges generate before it trickles into a data lake somewhere. Intel and plenty of others reckon the CPE is the place to do that and that the people who suck data from devices will welcome a VM sitting on the networks edge to take care of business.
Software developers are going to like this because most already offer their wares as packaged virtual appliances. If x86s arrive in every home, they've got a whole new market to address.
Carriers and ISPs should like it, because they get a new chance at selling services.
Yes: a lot of ducks need to line up for this vision to become reality, but Intel has a good shot at making it happen.
The company is a past master at making life easy for kit-makers, so it probably won't be long before it creates x86-based reference designs that modem-makers can use to crank out new products. Or it can bring them to market itself through Lantiq.
Oracle and HP mean the ideas outlined above should get a hearing in the corridors of big carriers and ISPs. And software vendors seldom sneer at a new channel.
Perhaps most importantly, there's an inflection point coming up as carriers deploy G.fast and DOCSIS 3.1 to cash in on punters' desire for gigabit speeds. Those upgrades won't thrive without a new CPE to take advantage of the new carriage standards. If Intel can help to create more capable CPEs, it gives carriers turning on G.Fast or DOCSIS 3.1 a way to encourage customers of laggard ISPs to churn. Winning new customers with a better CPE might pay for itself without ever having to sell services.
Of course carriers and ISPs have a long history of flubbing value-added services and being killed by over-the-top operators. This time around, however, there are powerful players wishing them to succeed. Because if carriers succeed, the likes of HP and Oracle win too and develop the kind of deep incumbencies that are hellish to unravel.
Intel's not saying all of the above will happen, but folk there have pointed out the scenarios outlined above as eminently possible and/or desirable. And if it does pan out, what's not to like: a multi-core server disguised as a modem/router can't be a bad thing to have in the home!
Market Data 
Markets 
AI
