Data Scientist Manager, Azure Network Engineering Group, Microsoft
Dr. Mehdi is founder of Network Data Science group within Global Networking Services at Microsoft. His team provided decision support services and operations research/ risk analysis to the shared cloud infrastructure. He studied Operations Research at MIT; subsequently he moved across the river where he was awarded doctorate from Boston University for his thesis on “Online and Offline Binpacking Algorithms for Resource Allocation under Extreme Uncertainty”. He provided analytical services to cloud services company Intralinks, Network Optimization Company LlamaSoft, Wearable startup Quanttus before joining Microsoft.
Currently he is Data Scientist Manager for the Azure Network Engineering group.
Roundtable: The Intersection of IoT, Cloud and Big Data
Sunday, 17 January 2016
The capacity of global telecommunication networks and its complexity has been exponentially growing. Traditional voice networks and internet service providers were both designed to deliver traffic between customers. Increasingly, enterprise businesses build their own network backbone to augment a content delivery platform, and employ network as a service. This change in purpose has transferred the way bits are routed within and across individual networks. The routing of traffic has been managed through traffic matrices. Intelligent traffic routing enables an intelligent network; one used to mitigate congestion, distribute compute and storage, enable reliability and resiliency, add a layer of protection and provides an ultimate cloud user experience.
As global networks today exist in an eco-system of 60,000-70,000 autonomous networks, the ability to control the flow of traffic within and across the eco-system is critical. To compute these multivariate traffic matrices in time efficient manner is imperative to cloud’s evolution. Tools like Label Switching Paths (LSPs) and flow like implementations are fast loosing relevance with an exponential growth of networks. Network topology morphs because of the growth and complexity pressures; this changes the utilization and resilience implications.
With most recent traffic growth CAGR in the three digit range, adequate planning is must to ensure that networks can be run at peak utilization. We extend research done by traditional telecom networks to enhance optimization algorithms. We have considered the optimal topology heuristics as a coupled problem, where we construct a global optimal solution which use the network as a sensor while predicting growth across cloud infrastructure and an actuator to manage the exponential growth trajectory. We present a global optimization platform that outperforms off-the-shelf bespoke solutions by over 50% while reducing the run time by half.