Jorge Arturo Cobb

Preserving quality of service guarantees in spite of flow aggregation

We investigate how quality of service may be guaranteed to a flow of packets in the presence of flow aggregation. For efficiency, multiple flows, known as the constituent flows, are merged together resulting in a single aggregate flow. After the network node where the aggregation occurs, packet schedulers are aware of the aggregate flow, but are unaware of its constituent flows. In spite of this, we show that quality of service may be guaranteed to the constituent flows, provided the aggregation is performed fairly. When the delay bound of a flow is de-coupled from the flows reserved rate, flow aggregation preserves the delay bound. When the delay bound of a flow is coupled to the flows reserved rate, flow aggregation preserves, and in some cases improves, the delay bound.

Time-shift scheduling—fair scheduling of flows in high-speed networks

We present a scheduling protocol, called time-shift scheduling, to forward packets from multiple input flows to a single output channel. Each input flow is guaranteed a predetermined packet rate and an upper bound on packet delay. The protocol is an improvement over existing protocols because it satisfies the properties of rate-proportional delay, fairness, and efficiency, while existing protocols fail to satisfy at least one of these properties. In time-shift scheduling each flow is assigned an increasing timestamp, and the packet chosen for transmission is taken from the flow with the least timestamp. The protocol features the novel technique of time shifting, in which the schedulers real-time clock is adjusted to prevent flow timestamps from increasing faster than the real-time clock. This bounds the difference between any pair of flow timestamps, thus ensuring the fair scheduling of flows.

Flow theory

We develop a simple theory of flows to study the flow of data in real-time computing networks. Flow theory is based on discrete and nondeterministic mathematics, rather than the customary continuous or probabilistic mathematics. The theory features two types of flows: smooth and uniform, and eight types of flow operators. We prove that, if the input flow to any of these operators is smooth or uniform, then both the internal buffer and delay of that operator are bounded. Linear networks of flow operators are introduced, and their internal buffers and delays are derived from the internal buffers and delays of their constituent operators. We extend flow theory so that it can be used in analyzing cyclic networks and networks of multiflows. Since many rate-reservation protocols can be represented as linear networks of flow operators, we use flow theory to prove that a number of these protocols (stop-and-go, hierarchical round-robin, weighted fair queueing, self-clocking fair queueing, and virtual clock) require bounded buffering and introduce bounded delay.

Congestion or corruption? A strategy for efficient wireless TCP sessions

We present a new acknowledgment strategy to improve the performance of TCP sessions that originate or terminate in noisy wireless networks for mobile computers. This acknowledgment strategy allows the TCP source to distinguish between losses due to congestion and losses due to corruption. With this distinction, the source can reduce its sending rate when congestion occurs, and quickly retransmit when corruption occurs. Without this distinction, TCP throughput is shown to suffer significantly over a path with a large bandwidth-delay product. The strategy is also appropriate for dealing with losses due to hand-offs of a mobile computer from one wireless cell to another.

A stabilizing solution to the stable path problem

The stable path problem is an abstraction of the basic functionality of the Internets BGP routing protocol. This abstraction has received considerable attention, due to the instabilities observed in BGP. In this abstraction, each process informs its neighboring processes of its current path to the destination. From the paths received from its neighbors, each process chooses the best path according to some locally chosen routing policy. However, since routing policies are chosen locally, conflicts may occur between processes, resulting in unstable behavior. Current solutions either require expensive path histories, or prevent processes from locally choosing their routing policy. In this paper, we present a solution with small overhead, and furthermore, each process has the freedom to choose any routing policy. However, to avoid instabilities, each process is restricted to choose a path that is consistent with the current paths of its descendants on the routing tree. This is enforced through diffusing computations. Furthermore, our solution is stabilizing, and thus, recovers automatically from transient faults.

Stabilization of General Loop-Free Routing

We present a protocol for maintaining a spanning tree that is maximal with respect to any given (bounded and monotonic) routing metric. This protocol has two interesting adaptive properties. First, the protocol is stabilizing: starting from any state, the protocol stabilizes to a state where a maximal tree is present. Second, the protocol is loop-free: starting from any state where a spanning tree is present, the protocol stabilizes, without forming any loops, to a state where a maximal tree is present. The stabilization time of this protocol is O(n deg), where n is the number of nodes, and deg is the node degree in the network.

Look-Ahead Routing and Message Scheduling in Delay-Tolerant Networks

Routing is one of the most challenging development issues in Delay-Tolerant Networks (DTNs) because of lack of continuous connection. Existing routing schemes for DTNs provide best effort service, but are unable to optimize QoS and support message priority. In this paper, we present a Look-Ahead Routing and Message Scheduling approach (ALARMS) which exploits more accurate knowledge about various parameters regarding routing to achieve better QoS in the DTN. We assume a variation of the well-known ferry model, in which there are ferry nodes moving along pre-defined routes to exchange messages with the gateway node of each region on the route and also pass to the gateway nodes look-ahead routing information about when it will arrive at each gateway node on the route in the next two rounds and how long it will stay. The gateway nodes use this information to estimate the delivery delay of each message when being delivered by different ferries, and schedule the message to be delivered by the ferry which arrives earliest at the destination. Simulation results show that ALARMS outperforms three existing routing protocols: epidemic routing, Spray-and-Wait, and Spray-and-Focus, in terms of delay time, delivery ratio, and overhead. We also discuss five enhancement strategies on ALARMS and how ALARMS can support message prioritization.

Enforcing convergence in inter-domain routing

The stable-paths problem is an abstraction of the basic functionality of the Internets BGP (border gateway protocol) routing protocol. This abstraction has received considerable attention, due to the instabilities observed in BGP. In this abstraction, each node informs its neighboring nodes of its current path to the destination node. From the paths received from its neighbors, each node chooses the best path according to some locally chosen routing policy. However, since routing policies are chosen locally, conflicts may occur between nodes, resulting in unstable behavior. Current solutions either require expensive path histories, or prevent nodes from locally choosing their routing policy. We present a solution with small overhead, and furthermore, each node has the freedom to choose any routing policy. However, to avoid instabilities, the possibility of divergence is measured using an efficient cost metric exchanged between nodes. If the cost metric indicates that divergence is occurring, steps are taken to ensure convergence.

Hierarchical-battery aware routing in wireless sensor networks

The Wireless Sensor Networks (WSN) have been envisioned to help in numerous monitoring applications. Pulsed battery discharge from the nodes prolongs the battery capacity as compared to constant discharge. Idle time between the pulses of discharge helps in recovering the discharged battery capacity. In the hierarchical routing protocol, cluster head nodes expend much higher energy than other nodes without any idle time. In this paper, we are proposing a Hierarchical-Battery Aware Routing (H-BAR) protocol for the WSNs. To make discharge from the nodes as pulsed as possible, role of cluster heads is changed periodically between the nodes. Battery recovery capacity depends on the remaining capacities of the battery. Nodes with the higher remaining capacities will have a higher probability of recoveringthan the nodes with the lower remaining capacities. Hence, to improve the recovery capacity, discharge from each node should be as uniform as possible. In the previous hierarchical routing protocols, such as LEACH, periodic selection of cluster heads is done probabilistically. Probabilistic election does not provide uniform discharge from the nodes. In the H-BAR, to provide uniform discharge, nodes with the higher remaining capacities are chosen as the cluster heads. Simulations show that H-BAR can improve the lifetime of the WSN up to 3 times over the lifetime of the WSN using the LEACH protocol.

Stabilization of Routing in Directed Networks

Routing messages in a network is often based on the assumption that each link, and so each path, in the network is bidirectional. The two directions of a path are employed in routing messages as follows. One direction is used by the nodes in the path to forward messages to their destination at the end of the path, and the other direction is used by the destination to inform the nodes in the path that this path does lead to the destination. Clearly, routing messages is more difficult in directed networks where links are unidirectional. (Examples of such networks are mobile ad-hoc networks and satellite networks.) In this paper, we present the first stabilizing protocol for routing messages in directed networks. We keep our presentation manageable by dividing it into three (relatively simple) steps. In the first step, we develop an arbitrary directed network where each node broadcasts to every reachable node in the network. In the second step, we enhance the network such that each node broadcasts its shortest distance to the destination. In the third step, we enhance the network further such that each node can determine its best neighbor for reaching the destination.