Navid Ehsan

Evaluation of performance enhancing proxies in internet over satellite

Performance enhancing proxies (PEPs) are widely used to improve the performance of TCP over high delay-bandwidth product links and links with high error probability. In this paper we analyse the performance of using TCP connection splitting in combination with web caching via traces obtained from a commercial satellite system. We examine the resulting performance gain under different scenarios, including the effect of caching, congestion, random loss and file sizes. We show, via analysing our measurements, that the performance gain from using splitting is highly sensitive to random losses and the number of simultaneous connections, and that such sensitivity is alleviated by caching. On the other hand, the use of a splitting proxy enhances the value of web caching in that cache hits result in much more significant performance improvement over cache misses when TCP splitting is used. We also compare the performance of using different versions of HTTP in such a system. Copyright

Modeling TCP performance with proxies

This paper investigates TCP dynamics and performance over proxies that shorten the TCP feedback loop by segmenting the end-to-end connection. Such proxies are often used to improve the TCP performance, e.g. a splitting/spoofing proxy in satellite communication, and more commonly, a web cache. Through analysis, we attempt to develop a basic understanding of the properties of TCP dynamics when such proxies are used, and further obtain design principles of systems involving such proxies. We present simple models capturing some features of the proxy performance in both the lossless and lossy scenarios. Due to the complexity involved, detailed analysis is only available in the lossless scenario, and our discussion in the lossy scenario is largely limited to steady-state behavior. Nevertheless we are able to obtain useful insight. We identify conditions under which using a proxy provides significant or marginal performance gain by investigating factors including initial window size, congestion level of the proxy, and the level of asymmetry between the segments segregated by the proxy. We also discuss how these conditions affect the deployment and provisioning of such systems.

On the optimality of an index policy for bandwidth allocation with delayed state observation and differentiated services

In this paper we study the optimality of an index policy for a bandwidth allocation problem, where a single server is allocated among N queues in a slotted system based on the queue backlog information. Due to the physical nature of the system this information is delayed, in that when the allocation decision is made, the server only has the backlog information from an earlier time. This results in imperfect and partial state observation. Queues have Bernoulli arrival processes with different probabilities of arrival, as well as different buffering/holding costs to differentiate heterogeneous classes of traffic/service. The objective is to minimize the expected total discounted holding cost over a finite or infinite horizon. We introduce an index policy with indices defined as functions of the state of a queue. We first show that when the state of the system is away from the boundary, i.e., no empty queues, the index policy is optimal. When there are empty queues, we show that under sufficient separation of the indices the index policy is still optimal. We show by example that if the separation does not hold, the index policy is not necessarily optimal. We then formulate the optimal bandwidth allocation as a restless bandit problem and show under what conditions the index policy calculated using Whittles heuristics, which in general is only asymptotically optimal, is optimal for the finite case

Optimal bandwidth allocation in a delay channel

In this paper, we consider the problem of allocating bandwidth to two queues with arbitrary arrival processes, so as to minimize the total expected packet holding cost over a finite or infinite horizon. Bandwidth is in the form of time slots in a time-division multiple-access schedule. Allocation decisions are made based on one-step delayed queue backlog information. In addition, the allocation is done in batches, in that a queue can be assigned any number of slots not exceeding the total number in a batch. We show for a two queue system that if the holding cost as a function of the packet backlog in the system is nondecreasing, supermodular, and superconvex, then: 1) the value function at each slot will also satisfy these properties; 2) the optimal policy for assigning a single slot is of the threshold type; and 3) optimally allocating M slots at a time can be achieved by repeatedly using a policy that assigns each slot optimally given the previous allocations. Thus, the problem of finding the optimal allocation strategy for a batch of slots reduces to that of optimally allocating a single slot, which is conceptually much easier to obtain. These results are applied to the case of linear and equal holding costs, and we also present a special case where the above results extend to more than two queues

Analysis of TCP transient behavior and its effect on file transfer latency

In this paper we present a Markov chain for TCP congestion avoidance phase. With this model we are able to analyze congestion window behavior as a discrete-time stochastic process and distinguish between window transient period and steady state. Using this result we are able to obtain more accurate estimate of TCP latency over lossy links compared to existing models. We then simplify the proposed model and show that the transient period evolves with an exponential rate. Our results are validated using NS2 simulation and show significant improvement in latency estimate for a wide range of file sizes.

Server allocation with delayed state observation: Sufficient conditions for the optimality of an index policy

In this paper we study an optimal server allocation problem, where a single server is shared among multiple queues based on the queue backlog information. Due to the physical nature of the system this information is delayed, in that when the allocation decision is made, the server only has the backlog information from an earlier time. Queues have different arrival processes as well as different buffering/holding costs. The objective is to minimize the expected total discounted holding cost over a finite or infinite horizon. We introduce an index policy where the index of a queue is a function of the state of the queue. Our primary interest is to characterize conditions under which this index policy is optimal. We present a fairly general method bounding the reward of serving one queue instead of another. Using this result, sufficient conditions on the optimality of the index policy can be derived for a variety of arrival processes and packet holding costs. These conditions are in general in the form of sufficient separation among indices, and they characterize the part of the state space where the index policy is optimal. We provide examples and derive the indices and illustrate the region where the index policy is optimal.

Dynamic Bandwidth Allocation for Low Power Devices With Random Connectivity

In this paper we consider the bandwidth allocation problem where multiple low power wireless devices share a common time-slotted channel for transmitting to a single server. Due to energy constraints, these devices alternate between common active and inactive periods, the former typically much smaller than the latter. At the beginning of each active period the server decides which user(s) can access the common channel. This decision is based on the knowledge of the current backlog and connectivity of each queue. In each time slot an active user may or may not be connected to the server. If a user is connected to the server, it can transmit with a certain success probability. Arrivals are arbitrary and there is a cost for holding a packet in the queue. Different queues have different packet holding costs leading to differentiated services. We consider the problem of minimizing the total discounted cost over a finite or infinite horizon and provide sufficient conditions under which a greedy policy is optimal. We consider two connectivity models: (1) there is no information about connectivity statistics, and (2) connectivity probability is independent from one time slot to the other (memoryless channel). We show that in each of these cases it is optimal to serve the user with the highest one step reward (smallest one step cost) if this gain is sufficiently larger than that from serving the other users. The sufficient condition is shown to be asymptotically tight in special cases.

Properties of optimal resource sharing in a delay channel

In this paper we consider the problem of allocating bandwidth/server to two user transmitters/queues with arbitrary arrival processes, to minimize the total expected holding cost of backlogged packets in the system over a finite horizon. However, the queue backlog information is delayed due to communication delay in the channel. In addition, the bandwidth allocation is done in batches, so that a queue can be assigned any number of slots not exceeding the total number in a batch. This problem is motivated by channel allocation in a communication system involving large propagation delay, e.g., a typical satellite data communication scenario. Our principal interest in this paper is to investigate whether the optimal assignment of a batch of slots can be achieved by sequentially using a strategy that is optimal in assigning a single slot, which is typically much easier to find. In this paper we show that if the cost c(x), as a function of the packet backlog x in the system, is non-decreasing, supermodular and superconvex, then (1) the value function at each time slot will also satisfy these properties; (2) the optimal policy for assigning a single slot is of the threshold type; and (3) optimally allocating M slots at a time can be achieved by repeatedly using a policy that assigns each slot optimally given the previous allocations.