IEEE Future Directions Talks 5G: Kaniz Mahdi - Part 2 of 2
Kaniz Mahdi is Vice President of Advanced Architectures at Ciena, responsible for Ciena’s global technology strategy and target architecture for 5G era networks underpinned with Open Architectures, Network Function Virtualization, Software Defined Networking, Machine Intelligence and Autonomics. She contributes to ongoing activities in IEEE SDN and IEEE 5G where her focus is setting up a roadmap for the evolution of the industry for 5G and beyond.
Question: Does network slicing fit into this design?, continued
Mahdi: Mobile networks, and this is a little bit of a history lesson here, mobile packet data systems were basically built using multitude of purpose-built Mobility and Security gateways bolted on a packet network originally designed for non-differentiated web browsing. This results in extra cost (in terms of resource consumption, as well as latency budget) and duplication of functionality as user packets traverse multiple functional stacks at different touch points on their path through the mobile system; not to mention complexity in configuring such systems, and operational inefficiencies resulting from layers and layers of features bolted on over time.
Current industry efforts around separation of Control Plane and User Plane functions with SDN/NFV to enable data plane programmability is an important first step, however, significant effort is needed to evolve current networks toward composability of lean systems that maximize extensibility with minimal redundancy, and completely eliminate software and hardware dependencies.
Ideally, it should be possible to replace such monolith gateways with Control Applications that drive application specific control behaviors through south bound API exerted on common data plane fabric. Stitching a user-flow then becomes simply a matter of stitching various control applications with east-west and north-south interfaces
Question: What are other challenges in implementing dynamic compositions?
Mahdi: Another fundamental challenge in implementation of dynamic compositions is placement, or I should say optimal placement of desegregated components. And there are two key factors. One, of course, is the practicality associated with the real estate; but I'm not going to go into the real estate issues here. Technology-wise, one of the key elements that needs to be accounted for is the demand-side equations that will dictate the network topologies and traffic distribution. When you look at current traffic topologies, current networks are basically configured with traffic flows that follow a very clear User-to-Content construct, which builds on centralization of data in large data centers with provision for improved efficiency in content delivery through internet access points and cache locations placed closer to the points of content consumption. Although this is a perfectly sound model for content delivery with effective use of capacity over metro and long-distance backbones, and low latency between users and content processing locations, it is challenged by ultra-low latency control expected with 5G, and order of magnitude higher volumes of latency constrained data exchange expected with IoT; in terms of both nodal interconnection and computational positioning.
Now, there has been quite a lot of advancement in this area, and most notably the central office re-architected as a data center / CORD initiative of ONF that has started to look into the transformation of Telco Service Providers Central Offices into Data Centers. This is an important first step in distribution of content and services across the time and space continuum that is needed for 5G and IoT. However, this level of distribution going all the way to the central office is not deep enough. There is a deeper level of distribution that is needed to serve the high volumes of low latency data exchange expected with 5G, specifically, URLLC, the ultra-reliable, low latency communication.
What you need is, again, I alluded to this earlier, a system which is a based on an Intelligence hierarchy, a system that distributes Intelligence across a time and space continuum, with control loops staggered across the two extremes represented by the ‘user-device’ and the ‘point of content origination’.
This may very well be the case of easier said than done. Staggering of control loops is quite a complex problem and that brings me to the toughest challenge that we face in the real-life implementation of the hierarchical systems, and that is optimal distribution intelligence.
Question: What are the challenges in distributing intelligence across the time and space continuum?
Mahdi: One of the key enablers that I see that would have to be in place for us to optimally distribute control (hence intelligence) is something that has been studied for decades but has just recently shown some real-world implementation promise, and that is Autonomics.
Autonomics, in the context of automated network management has been researched in academia with selective participation from industry for over a decade now. Just recently, a handful of industry initiatives have started to look at the practicalities around use of autonomics for automated network management. And this is still a very new, very nascent area of research and there's lots of open research questions. I'll name a few, as examples, but this is not an exhaustive list:
Traffic modeling of device swarms expected with newly defined or yet to be defined autonomous processes; for example, self-driving cars. Traditionally, traffic modeling practices have centered around human-to-human communications, and they’re based on years and years of data collected from operational systems. But the advent of Self-Driving Everything, it's hard to predict a specific pattern or the patterns that could emerge – how these services will end up shaping the future traffic patterns? That needs to be modeled somehow.
Mobility management is another key issue. Mobility management schemes that exist in the mobile network today are all based on centralized anchoring of services. What we're doing with the distribution of control is reversing that model. We are distributing a control system across multiple steps in the mobile topology, and that creates a completely new problem that I imagine would need some novel techniques for solving the mobility problem.
Data ownership and federation across multiple controlled jurisdictions is yet another key problem to be solved. This would be the case when a service flow requires stitching of heterogeneous control systems owned and operated by disparate Service Providers. For example, a home automation appliance may have to be stitched to an automated street surveillance system that may in turn be stitched with a Central Office that is controlling the neighborhood and then eventually that would have to be stitched with the central controller management of the communications network, which is stitching all of these together. I just basically walked you through three or four different jurisdictions, and all of these could in practice be implemented with their control and management and automation algorithms using their own data schemes. All of these data schemes would have to be brought together somehow. That's a real-world implementation problem that needs to be solved.
These are just a few examples off the top of my head, one could think of many more.
Q: What are we doing to solve these problems?
Mahdi: Let's look at the state of the industry.
We see Access, Cloud and IoT market segments converging toward what’s represented by a broad suite of technologies ranging from Compute and Intelligence capabilities residing on a mobile user device (e.g. vehicle, or handset); located in a home (e.g. home automation appliance); or an enterprise (e.g. local service network); or positioned in the network at a Cell Tower, or a Central Office.
Several industry initiatives, currently underway, are exploring and/or developing different facets of this technology suite as driven by the specific needs of the originating market segment.
The IoT folks have been working on Open Fog for a while, and I should say (and this is just my personal opinion) that Open Fog is closer to the target space that will end up being implemented, which is basically distribution of compute and intelligence across a time and space continuum that extends between content origination all the way to the point of consumption.
The telecommunications industry, at this point, is consumed with reshaping the access network. Basically, a lot of effort is taking place around implementation and verification of 5G and the other access technologies that would have to be factored into 5G for implementation of Multi-Access networks. The telecommunications industry has ways to go before they get to a point where they could shake hands with (hopefully) the folks in Open Fog who are designing the continuum.
As I alluded to earlier, the industry seems fragmented at this point, which is natural, considering we’re in early infancy of the technology and the fact that multiple industries are involved. We do acknowledge the need for multiple reference implementations that would cater to different market segments. However, interoperability across these disparate implementations is paramount to ubiquitous provisioning of services across this continuum, which may often times transcend multiple jurisdictions. So how should we solve this problem? Historically, when people have attempted to solve such issues, they were trying to come up with an all-encompassing standard that unifies all potential domains and jurisdictions that could be involved. But in my personal experience, I have not seen previous attempts to solve similar problems with an umbrella standard as being fruitful or effective. What one could hope for, or a call to action to see if we can make it happen, is perhaps a common architecture pattern that will stitch these disparate reference implementations using open API, open reference models, open implementation models and the right level of abstractions that are needed to stitch these reference implementations together, building on some common core principles. What comes to mind here is the SDN architecture that has been set forth by ONF, and to my knowledge it has been adopted universally by all industries so far. So that could be a good starting point to bring the industry together to help resolve some of these challenges we face in implementing practical systems that will help us fulfill the promise of 5G and SDN.