Last week I spoke at a small workshop about how artificial intelligence can be used in multi-layer network planning, optimisation and fulfilment. Multi-layer networks are causing service providers a bit of a headache at the moment. Typically they comprise MPLS or Ethernet over a DWDM optical transport network. On paper it looks quite neat: Very high capacity transport that can carry pretty much any traffic type, and super-flexible IP services on top.
It is generally agreed that the effort is worth-while in order to offer high bandwidth services and provide a platform for rapid introduction of new types of service. But the PowerPoint diagrams don’t translate to the real world so easily.
A few of the problems service providers face are:
- Migrating legacy services is complex.
- There’s a wide choice of DWDM deployment options, each with their own pros and cons.
- Most established planning systems are unable to design services across more than one network layer at once.
Most of the presenters at the workshop were addressing the last point. I had thought, by talking about artificial intelligence, I would be providing a unique ‘intellectually rigorous’ approach to solving the problem.
Wrong.
While I was the only one talking about A.I. in general, all but one of the OSS vendors attending were pitching ‘intelligent’ solution. And all of those were start-ups, including Aria Networks who I was representing.
So it seems intelligent planning is an emerging solution in OSS. Certainly amongst this group of hardware & software vendors and service providers, there was a clear expectation that intelligence is essential to enable effective, optimal planning. But few approaches are actually well understood, and certainly no one approach has emerged as a clear leader.
What do I mean by intelligent? In researching my presentation I found no established definition that satisfied the problems and needs of telco. So, to quote myself:
“Applying better algorithms and better analysis to out-perform the capabilities of scripted, procedural, and BPEL approaches.” – Me (Last Week)
For network planning, the core requirement is optimal routing (or re-routing) of large numbers of circuits and services. Procedural approaches don’t scale simply because the search space is too big: Too many services; too many devices; too many links; to many network layers; too many parameters; too many constraints. You can’t crack this with brute force.
Intelligent planning has to find a way to reduce the search-space, balance all the requirements, while still being sure that the optimum solution is found. Various technologies, algorithms and approaches exist, such as:
- Neural nets
- Multiple integer programming and linear programming
- Genetic evolution
- Heuristic-driven path finding
The challenge facing OSS vendors and service providers is a lack of understanding of non-deterministic A.I. techniques. Much A.I. is well proven in other fields, yet use in OSS has been limited until recently. There’s a huge gap between peoples understanding of a procedural solution (a number of tasks processed from start to finish) and a non-deterministic approach which is beyond the understanding of most users (me included usually). Yet it is only non-deterministic intelligent solutions that will be able to continue to optimally manage capital expenditure in multi-layer next-generation networks.
So, we have some work to do. Like inventory ten years ago, and service catalogs three years ago, it turns out that OSS marketing is all about education again.
