Layering as Optimization Decomposition: A Mathematical Theory of Network Architectures
read more
Citations
Distributed Subgradient Methods for Multi-Agent Optimization
Controllability of complex networks
Fog and IoT: An Overview of Research Opportunities
Multilayer Networks
Stochastic Network Optimization with Application to Communication and Queueing Systems
References
A mathematical theory of communication
Convex Optimization
Random early detection gateways for congestion avoidance
Congestion avoidance and control
Rate control for communication networks: shadow prices, proportional fairness and stability
Related Papers (5)
Frequently Asked Questions (11)
Q2. What are the future works in "Layering as optimization decomposition: a mathematical theory of network architectures" ?
This section highlights some of the main issues for future research and their recent developments.
Q3. What is the way to solve the coupling problem in utilities?
One way to tackle the coupling problem in the utilities is to introduce auxiliary variables and additional equality constraints, thus transferring the coupling in the objective function to coupling in the constraints, which can be decoupled by dual decomposition and solved by introducing additional consistency pricing.
Q4. What are the main metrics to be considered when comparing distributed algorithms?
To compare a variety of distributed algorithms, the following metrics all need to be considered: speed of convergence, the amount and symmetry of message passing for global communication, the distribution of local computational load, robustness to errors, failures, or network dynamics, the impact to performance metrics not directly incorporated into the objective function (e.g., userperceived delay in throughput-based utility maximization formulations), the possibility of efficient relaxations and simple heuristics, and the ability to remain evolvable as the application needs change over time.
Q5. What is the engineering implication of the dual-decomposition-based distributed algorithm?
The engineering implication is that appropriate provisioning of link capacities will ensure global convergence of the dual-decompositionbased distributed algorithm even when user utility functions are nonconcave.
Q6. How can a dynamic link cost be stabilized?
Routing can be stabilized by including a strictly positive traffic-insensitive component in the link cost, in addition to congestion price.
Q7. What are the two possible policies for rate-reliability provisioning?
In the case where the rate-reliability tradeoff is controlled through the code rate of each source on each link, there are two possible policies: integrated dynamic reliability policy and differentiated dynamic reliability policy.
Q8. What is the main reason why TCP Reno is a performance bottleneck?
It is also well-known, however, that as bandwidth-delay product continues to grow, TCP Reno will eventually become a performance bottleneck itself.
Q9. What is the persistence probability of link l?
After each transmission attempt, if the transmission is successful without collisions, then link l sets its persistence probability to be its maximum value pmaxl .
Q10. What is the simplest way to show that x can depend on the flow arrival pattern?
The authors now present a simulation using Network Simulator 2 (ns2) that shows that x can depend on the flow arrival pattern because of the existence of multiple equilibria.
Q11. What is the signal-to-interference ratio for a given set of path losses?
The signal-to-interference ratio forlink l is defined as SIRlðPÞ ¼ PlGll=ð Pk 6¼l PkGlk þ nlÞ for a given set of path losses Glk (from the transmitter onVol. 95, No. 1, January 2007 | Proceedings of the IEEE 283logical link k to the receiver on logical link l) and a given set of noises nl (for the receiver on logical link l).