NFV Specs/Open Source

BT Threatens to Ditch OpenStack

DUSSELDORF -- SDN & Openflow World Congress -- BT is threatening to abandon OpenStack in favor of a proprietary technology during its rollout of virtual enterprise services unless vendors backing the standard can overcome six potential technology showstoppers.

Such a move by the UK fixed-line incumbent could be a major setback for OpenStack, an open standard that is being touted as one of the key technologies underpinning NFV.

BT Group plc (NYSE: BT; London: BTA)'s concern relates to the use of OpenStack with virtual enterprise CPEs, whereby servers are distributed around the edge of the network and on customer premises.

Peter Willis, BT's chief researcher for data networks, acknowledges that OpenStack has major attractions as set of open source tools, with backing from some of the world's biggest IT companies, but says the technology still holds a number of risks for an operator rolling out virtual enterprise CPEs.

Unless the OpenStack community can address a number of serious challenges, BT will look to use an alternative technology.

Those challenges relate to the connection of virtual network functions (VNFs), service chain modification, scalability, so-called "start-up storms," the security of OpenStack over the Internet and backwards compatibility.

"If these six issues are not addressed we will not use OpenStack for virtual enterprise," he told attendees here in Dusseldorf. "There are proprietary solutions out there we could use and we will use them in lieu of OpenStack being available."

For more NFV-related coverage and insights, check out our dedicated NFV content channel here on Light Reading.

Willis revealed that some of the most important OpenStack stakeholders -- including HP Inc. (NYSE: HPQ), Red Hat Inc. (NYSE: RHT), Canonical, Mirantis Inc. and Intel Corp. (Nasdaq: INTC) -- have agreed there are problems but have not yet reached a consensus on how to address them.

Although BT could opt for a proprietary technology in the short term and introduce OpenStack at a later stage, this will become more difficult as services are rolled out.

"It's a strategic migration issue," says Willis. "It's easy to start off greenfield because we've not deployed any large numbers of virtual enterprise CPEs but in two or three years there will be more legacy."

Next page: The six challenges

1 of 2
Next Page
Page 1 / 2   >   >>
kq4ym 10/26/2015 | 9:48:02 PM
Re: Rather ridiculous I'm sceptical that BT would follow through with the "threat." I'm just guessing they;re bluffing in an attempt to speed up the openstack standards solutions. Going to proprietery is much like moving backwards at this point.
ahmadarshad 10/16/2015 | 10:21:33 AM
Resistance is one form of acceptance I take it a positve step. Risks pointed out in this articles shoud be taken positively. Openstack is still in its infancy and there are fears and uncertaintity everywhere on its use and deployment. Best thing is to contribute and be part of it as pointed by EtherealMind. I like see these type of comments/risks from other Service Providers and users so that openstack community starts lokking in these areas. We face these challenges in our professional life every day and either get them resolved or find a work around. I believe security of infrastructure will be a big challenge in coming years and will be a hot topic of discussion.
Director72287 10/15/2015 | 7:11:31 PM

Overall this smells like a FUD piece designed to demonstrate that a major service provider is unlikely to adopt OpenStack at this point in time. For starters, the idea that OpenStack is suitable for a major telco service provider's  virtual enterprise CPE platform is implausible at best.

If (and that's a big IF) Peter Willis is actually considering it he's guilty of trying to bang a square peg into a round hole, he should know better. I don't believe he is, so I have to assume this piece is an indirect rebuttal to a number of vendors who are hammering on his door telling him he should be using OpenStack because it is the answer to a maidens prayer. OpenStack is not an SDN platform. OpenStack could be used to build a service platform alongside other components, but that's a different discussion.


To address the points directly...


VNF's. Virtual Network Functions, are not a core use-case that OpenStack was ever designed to address. It may evolve to do that, but right now you need a focussed SDN solution to address this one.


Service chaining, the connection of a number of VNF's in series. It's just an amplification of the previous point.


Lack of scalability with a single controller.  Spare me! No sane commercial, service provider, or enterprise grade implementation would try to run with a single controller, scaling OpenStack requires an understanding of how to scale distributed systems.


Start-up storms. Anyone who thinks they can load 100,000 virtual CPE's onto a single controller has either been at the pixie dust, or is lacking clue about scaling distributed systems, or both (see previous point). Solving the problem of high volume SSL session setup is a well trodden path, horizontal scaling is one possibility, doing it in hardware is another (see previous point, again)  


Security. Yes, there is work to be done in securing OpenStack. SSL gets us a long way down the path at the transport layer, but there are other aspects as well. To state that a hypervisor to controller connection requires "500 pinholes" in a firewall is naive demagoguery, you need about 8, and that could probably be reduced even further if you're really security conscious and prepared to tweak the control plane architecture.


Backwards compatability. This problem exists for every single piece of moderately complex software with lifetime of more than a few months. In the specific case of OpenStack it's more complex than most, OpenStack has a lot of moving parts, and is developed by a community of loosely coupled projects. Synchronising them all is a challenge. To adopt open source software into your platform and enterprise usefully you have to invest in understanding the community and contributing to it.


Disclosure: I'm a huge fan of OpenStack, I'm a director of an OpenStack platform operator. It's a great platform, but it's not the answer to every question.

Flames will be piped to /dev/null :-)

mendyk 10/15/2015 | 11:06:46 AM
Re: Rather ridiculous Heavy Reading's most recent research shows incredibly strong support for OpenStack among network operators, at least right now.
NeilMcRae 10/15/2015 | 3:34:51 AM
Re: Rather ridiculous I would say yes - it also, in many use cases, performs very well.
Steve Saunders 10/14/2015 | 5:12:59 PM
Re: Rather ridiculous Is VMware really carrier class code? I know they are now sniffing around the t4elecom market but i thought they had Enterprise pedigree. 
NeilMcRae 10/14/2015 | 4:57:48 PM
Re: Is this a common view? We always want more!  We want the best for our customers!

Openstack scalability is not a new issue and Peter is merely stating a fact and its not a threat its simply an ackowledgement of the current challenges operators face in rolling NFV (and other cloud services) out. Its not an issue of being competitive its an issue of having something that works for customers, at BT we only deploy things that work for customers!

There are alot of other gaps in Openstack and in time I am sure they will all be resolved, however with the current focus on these services customers are starting to want these services and the whole ecosystem is at varied levels of maturity; We are acknowledging an issue that we feel is not being as widely recognised as it needs to be and in many cases this is preventing large scale deployment. For vendors in the NFV/SDN space its going to be a barrier to sales for sure (I'd argue it already is).

Also its not the only issue in this space, lets not forget that this technology is still in its infancy, there are still significant OSS/BSS issues with SDN/NFV. The IT stack and capability to manage large scale SDN/NFV deployments and wide range of interacting services simply doesn't exist right now and how these technologies overlap with the existing telco and datacentre networks is still very "homebriew". For BT Cloud Connect we have created something to fill the gaps but its iimited in its capability, we want something more that truly brings abstracted service creation on any type of containerised/virtiualised/appliance based solution hence our significant contribution to OpenConfig and other Yang models across the network space. The end to end architecture in areas of service inventory, logical and physical inventory, workflow, activation and telemetry is still at a very early stage, with very little alignment or deployable solutions. Without these things, some of the promise these new technolgies bring won't materialise.

How does workflow work? how do atomic services interact with one another? how do we really drive the on demand opportunities between different methods of delivering services over different technologies and infrastructure - there is a long way to go.  If my customers fixed broadband Internet service goes down, how to I create a standby internet service using personal hotspot on my customers cellular service on demand whilst the fixed service is restored?

These are all maturity issues but we feel not enough focus is on this topic. We are all focused on interop and POCs, well frankly we've done that, infact we did it _years_ ago, now we want to move to proper deployment and for that we need a more complete ecosystem. But to be clear we aren't making any threats or otherwise we just want it to work for our customers and we will continue to collaborate to do this. And lightreadng if you want to come and have a chat about this, we are always open for coffee!




brooks7 10/14/2015 | 4:50:55 PM
Re: Rather ridiculous So, you can wait for OpenStack to fix their problems someday or implement today using VMware?  Is that the argument?

And by the way, all kinds of SaaS operators use VMware....and make money.  And I don't work for them, but I have been trying to raise this point for over a year.  Which is that SaaS operators have things pretty well solved already.


Steve Saunders 10/14/2015 | 3:39:20 PM
Re: Rather ridiculous i need to dig in a bit more - thans for the insights. 
EtherealMind 10/14/2015 | 3:34:55 PM
Re: Rather ridiculous The issues raised are well known, raised several times by others and OpenStack has been working on them for about a year and making progress. It might take a couple more years to get stable code and viable delivery. If BT got involved, comitted funding and developers then it could happen faster. Instead of gum flapping, they should participate. 

The only alternative that does have solutions for the issues raised is VMware and comes at a licending between £200-800 per physical server or ~£50-100 per VM (depending). 

There are OpenStack solutions from Cisco, HP, Mirantis and others that offer some answers but, again, cost per VM would make BT uncompetitive. 

Also I'm not raising the fact that OpenStack isn't specifically intended to handle lagacy applications that exist in private data centers. 

Page 1 / 2   >   >>
Sign In