AT&T: Sorry vendors, SDN is eating your lunch

In news that will chill purveyors of big networking iron, AT&T last week told its earnings call it reckons its software-defined network (SDN) rollout will cut its capital expenditure. While reports last week focussed on the company's revenue (US$33 billion for the second quarter), margins (36.3 per cent), and operating income …

[Snow Wombat]

They sure as hell aren't laying fibre.

>capex is already slimming down from $20 billion in 2014 to “around” $18 billion predicted for this year, and that includes capex devoted to AT&T's expansion efforts in Mexico.

So despite them being able to gobble Dish Networks, and there being a "lay fiber to 12.6 million locations" conditions being attached to that deal, they are actually winding DOWN capex and build expenditures.

It's ALMOST as if they have absolutely no plans to actually build that fiber to meet those conditions.........

Naah.... I am just mistaken right?

[JetSetJim]

Re: They sure as hell aren't laying fibre.

The article is commenting on how SDN/NFV is giving an opportunity to not spend quite so much money on their mobile network as the expensive ATCA chassis/boards typically used to house the core functions like MSC, GSN, MME, SGW, etc... can be replaced by cheap "clouds" where each function is a program running on an Intel board, scaling up as demand ramps by dint of just adding another VM into the mix.

It's not a comment on their domestic broadband market, which doesn't have this opportunity.

Not that AT&T won't want to spend the least amount of money (and perhaps even a bit less) to meet (or narrowly miss in a manner that's able to be spun appropriately to make it seem like it's been met) their obligations.

[Anonymous Coward]

Bloodbath coming

I don't think the big tin-shifters quite appreciate how determined AT&T are to complete the move to SDN (and its cousin NFV). They are assuming, I think, that AT&T and similar big carriers will attempt it, fail for internal cultural reasons, and come back into the fold.

This isn't going to happen. Openstack may be wildly imperfect and virtualization may be difficult for many telco services, but the winds of change are at gale force. Time to get out of tin and into software.

[Anonymous Coward]

Yes and No

[AC as I am employed by one of the big tin-shifters and I am sure my views don't align with our formal communications]

I think that SDN/NFV have great value -- but major carriers will end up being selective how they use them. Some functions can, and will, certainly be virtualised -- and cost savings will be achieved. On the other hand, some network packet handling, and even some software functions, are still critical to handling massive data flows and keeping stability.

I have worked in this area for over 30 years and used to design and code high end routers. I have seen horrible behaviour of networks under stress once packets start being missed, control messages being delayed, etc. Although my technical skills are out of date, I am certain the problems have not got any easier.

I still believe SDN is mainly an Enterprise (not Carrier) technology and that NFV will be applied selectively. The article is correct that there is a trend for less intelligence in edge equipment but there is still a lot of value (i.e. cost for the carriers) in that gear. And that still requires expensive, specialised hardware (for example, for video, as well as just shifting packets around as fast as possible) and also bespoke software.

[Anonymous Coward]

Re: Yes and No

I'm mostly with you. Network performance in cloud is truly terrible - and no surprise, since openstack has a devil-may-care throw-bridges-at-it sort of implementation that makes it incredibly slow for packets to come off the wire and get to the VM. That really spoils your day if you live and die by processing tiny little RTP packets at wire speed! Say, if you are a telco...

Having said that, those problems will be overcome, and then it's only a matter of time before one of the majors does a sufficiently successful implementation that validates the market, and then everyone will want to try. I suspect the really func part will be convincing telco managers to build things completely differently. For the past 100 years they've been really good at specifying things to the n'th degree (erlangs and all that). How much of that learning gets chucked away?

Interesting times.

[martyvis]

Re: Yes and No

Marhinal Openstack network performance is why some vendor's approach to NFV is to use a carrier-grade solution. In HP's case, they are using a Wind River Linux kernel, as well as the Intel DPDK libraries. Shifts throughput from 2Gbps per VM to 20Gbps.