In recent years, there has been a critical rise in the utilization of low Earth orbit (LEO) satellite internet connectivity services. In certain regions, the growing demand outpaces the available data rates, thereby compromising future quality of service. While deploying more satellites is an acceptable short-term solution, a more durable alternative involves the design of higher-performance satellites with higher transmission capabilities. These next-generation payloads will require a very high capacity only achievable by inter-connecting multiple onboard processors to scale throughput significantly, in comparison to the traditional single-processor satellites. However, there is a lack of current research on the internal packet routing of high-throughput satellites between these onboard processors. To tackle this challenge, after modelling the internal satellite architecture, we design a real-time optimal flow allocation and priority queue scheduling method using online convex optimization-based model predictive control. We formulate the problem as a multi-commodity flow instance, where incoming data streams with distinct priorities are the commodities, the uplink beam is the source, and both the downlink beam and discarded packets are sinks. We use an online interior-point method to iteratively solve the routing and scheduling optimization problem, accounting for time-varying equality constraints as information about incoming flow is observed. Then, we feed back its decision in the satellite model. Our approach minimizes packet loss cost and facilitates real-time rerouting in response to incoming information and exogenous uncertainty. Our online convex optimization method provides low computational overhead and is implementation-ready. Our method is tested in a next-generation high-throughput satellite model and is compared to a reference batch optimization model and traditional methods such as cost-proportional allocation.

In recent years, deploying distributed applications efficiently over cloud platforms has been extensively studied. Various algorithms were developed to solve this optimization problem, such as virtual network embedding algorithms.

The HEAVEN platform is an embedded wireless datacenter solution that provides resilient and flexible internet coverage. This solution is an efficient way to provide emergency connectivity through to its mesh network. The next logical step after establishing connectivity is to provide processing power to the users. Exploiting drones' processing power can be a method for offloading computations, providing essential offline emergency services. With limited connectivity between devices, link allocation becomes a key component of the problem.

This talk presents a comprehensive analysis of the capabilities such platform could offer, by presenting a simulation based on individual device capability and connectivity. The simulation framework incorporates models for device capability, connectivity, and application requirements, enabling the evaluation of efficient application deployment strategies over HEAVEN. Optimizing application placement in real-time and in a distributed way would expand the network resiliency to its applications.

Additionally, this talk presents preliminary results on implementing various placement algorithm to the simulation framework, following the work presented in a previous conference, aiming to understand and solve this distributed virtual application placement problem under heavy network and resource constraints.

We study the problem of the optimal design of fiber-to-the-home optical access networks. Given a network of nodes and edges with a given demand for optical fibers at a subset of these nodes, the problem entails finding the optimal placement of splitters, which allows multiple demand points to share a common fiber between the central office and a splitter, such that the cost of fiber cables is reduced without incurring too much of splitter cost. Additionally, it needs to decide on the optimal selection of a cable type of appropriate capacity on each edge of the network to carry the required traffic. For this, we provide a mixed-integer programming (MIP) formulation of the problem and propose several valid inequalities (VIs), with or without a pre-specified template, to strengthen the formulation. Through our computational experiments, we demonstrate the efficacy of our proposed VIs, which help improve the lower bound of the problem from 79\% to 86.9\% of the MIP optimal cost, on average. For the special cases of the problem studied in the literature, we show that our formulation produces much tighter lower bounds. On top of that, our proposed VIs are comparatively much more effective in tightening the bounds. Specifically, our proposed formulation with our VIs consistently outperforms that available in the literature, being as much as 500 times faster in some instances.