The week trip to IFIP/IEEE International Symposium on Integrated Network Management (IM 2017) in Lisbon was fantastic. I had the chance to catch up with old friends and colleagues (Edmundo, Marilia, Alberto, etc.) and to meet other enthusiasts in network management, SDN, QoE, 5G, block chain and cognitive technologies.
I spent my first day with the QoE-Management workshop, which had one keynote led by seven presentations. There is a lot of work on measuring different aspects (delay, switching, fairness, buffer underrun) of the quality of adaptive streaming. Machine learning is also gaining its popularity in QoE management. In my opinion, the QoE communities face a few hurdles for a major leap ahead: human intent/perception, encrypted traffic, feasible machine learning solution in communication networks, and end-to-end multi-service management. I am glad to see that this community is very open to the challenges ahead. It is also quite interesting to see Tobias opening up the argument on Mean Opinion Score (MOS). MOS is essentially a method to gather and analyse user opinions in subjective experiments. MOS has been widely used in the QoE community for decades but it is mathematically flawed. I discussed this five years ago in a paper at IEEE CCNC: Statistical Analysis of Ordinal User Opinion Scores (Warning! It will upset you if you’ve done a lot of work using conventional MOS… If you ended up upset, seek doctor’s advice. Preferably a doctor in Mathematics.). Tactile Internet was mentioned a few times as one of the use cases. I think someone also mentioned NFV in user terminal with incentives? Why not…
The second day’s programme started with Raouf Boutaba (University of Waterloo)’s keynote on 5G network slicing. Raouf talked about virtual network embedding (VNE) with which we map virtual network nodes and links onto physical infrastructure. A good VNE would lead to better error tolerance, efficiency, and “collective wellbeing”, etc. It is surely linked to the cognitive networking that I am working on. Later on, a few papers from the industry dominated the experience track. Some highlights are Cisco’s model driven network analysis using a variation of RFC 7950 YANG (YANG is a data modelling language used to model configuration data, state data, Remote Procedure Calls, and notifications for network management protocols.); UNIFY, a framework that brings cross-layer “elasticity” that unifies cloud and service networks; virtualization of radio access networks (for end-to-end management and other purposes); and IBM’s “BlueWall”, an orchestration of firewalls. BlueWall still keeps human-in-the-loop so it’s probably more of an Intelligence Augmentation system rather than Artificial Intelligence. The Panel on “Challenges and Issues for 5G E2E Slicing and its Orchestration” was packed with good talks on 5G. People were very optimistic of 5G open slicing, especially its potential in creating future generation mobile operators (“anyone can be an operator”) and the E2E benefits on VR and emergency use cases.
The third day was led by two inspiring keynotes: “Intent-Driven Networks” from Laurent Ciavaglia, Nokia and “The Future of Management is Cognitive” from Nikos Anerousis, IBM Research. They recognised that network/service management is moving towards “dark room + algorithms” (machine learning), but human will still have pivotal roles: referring/curating knowledge and training systems to solve complex problems. I then went to the security session and SDN session for the rest of the day. An Ericsson talk discussed COMPA (Control, Orchestration, Management, Policy, and Analytics) adaptive control loop as an automation pattern for carrier networks, a good work to follow if you do such high-level designs. There was an interesting paper on addressing the shortage of scarce and expensive TCAM memory on SDN switches using “memory swap”. The idea is to employ the memory of SDN controller for least frequently used flow rules to free up TCAM space. Is it impractical, naive? I think there are scenarios where this solution will actually work well…
David Gorman from IBM kicked started the fourth day with his excellent keynote talk on “Making Blockchain Real for Business”. David shared his vision on a world of shared ledger, smart contract, privacy (certificate) and trust. He used auditing as one of the use cases to demonstrate the uniqueness of blockchain in tracking transactions (changes) in comparison to conventional database solutions. His talk then converged on a brief introduction of Hyperledger, a community effort on cross-industry blockchain technologies. I had a short and interesting discussion with David on the impact and use cases of blockchain in higher education. Ultimately, blockchain is merely a technology and not a solution (in fact, the same applies to SDN). I think it can be a key technology to enable cross-service end-to-end management but in many cases, a solution is not dictated by the technology but politics and regulations.
On the last day, I only stayed till lunch time before I had to leave to catch my flight. The highlight of the day is certainly Alex Galis (UCL)’s talk on Programmability, Softwarization and Management in 5G networking. He emphasised on the importance and impact of softwarization and network programmability, especially the quality of slice in future networks. I’d summarise his talk, blending in my own views, as autonomous, adaptive, and automated end-to-end resource management. Alex also spent a few slides concluding on the key challenges on network slicing, which are very helpful to new researchers in this field.
All in all, IM 2017 at Portugal has been a wonderful event (In fact, they’ve done so well that they also won Eurovision 2017). I am looking forward to its future iterations (NOMS and IM).