Amy Csizmar Dalal

Optimal scheduling in a queue with differentiated impatient users

We consider a M/M/1 queue in which the average reward for servicing a job is an exponentially decaying function of the jobs sojourn time. The maximum reward and mean service times of a job are i.i.d. and chosen from arbitrary distributions. The scheduler is assumed to know the maximum reward, service rate, and age of each job. We prove that the scheduling policy that maximizes average reward serves the customer with the highest product of potential reward and service rate.

Improving user-perceived performance at a World Wide Web server

We examine a method to improve the service of incoming requests at a World Wide Web server. The motivating factor is the existence of an impatient user pool: a user aborts a pending Web request if a response is not received within a random timeout period. We use a queueing theory approach to derive an optimal service ordering for this server, assuming Poisson arrivals and exponential service times. We find that the optimal policy is greedy, in that at any time the server processes the request with the highest perceived payoff. We verify these results both analytically and via simulation.

Towards Real-Time Stream Quality Prediction: Predicting Video Stream Quality from Partial Stream Information

While mechanisms exist to evaluate the user-perceived quality of video streamed over computer networks, there are few good mechanisms to do so in real time. In this paper, we evaluate the feasibility of predicting the stream quality of partial portions of a video stream based on either complete or incomplete information from previously rated streams. Using stream state information collected from an instrumented media player application and subjective stream quality ratings similar to the Mean Opinion Score, we determine whether a stream quality prediction algorithm utilizing dynamic time warping as a distance measure can rate partial streams with an accuracy on par with that achieved by the same predictor when rating full streams. We find that such a predictor can achieve comparable, and in some cases markedly better, accuracy over a wide range of possible partial stream portions, and that we can achieve this using portions of as little as ten seconds.

Discerning user-perceived media stream quality through application-layer measurements

The design of access networks for proper support of multimedia applications requires an understanding of how the conditions of the underlying network (packet loss and delays, for instance) affect the performance of a media stream. In particular, network congestion can affect the user-perceived quality of a media stream. By choosing metrics that indicate and/or predict the quality ranking that a user would assign to a media stream, we can deduce the performance of a media stream without polling users directly. We describe a measurement mechanism utilizing objective measurements taken from a media player application that strongly correlate with user rankings of stream quality. Experimental results demonstrate the viability of the chosen metrics as predictors or indicators of user quality rankings, and suggest a new mechanism for evaluating the present and future quality of a media stream.

User-perceived quality assessment of streaming media using reduced feature sets

While subjective measurements are the most natural for assessing the user-perceived quality of a media stream, there are issues with their scalability and their context accuracy. We explore techniques to select application-layer measurements, collected by an instrumented media player, that most accurately predict the subjective quality rating that a user would assign to a stream. We consider three feature subset selection techniques that reduce the number of features (measurements) under consideration to ones most relevant to user-perceived stream quality. Two of the three techniques mathematically consider stream characteristics when selecting measurements, while the third is based on observation. We apply the reduced feature sets to two nearest-neighbor algorithms for predicting user-perceived stream quality. Our results demonstrate that there are clear strategies for estimating the quality rating that work well in specific circumstances such as video-on-demand services. The results also demonstrate that neither of the mathematically-based feature subset selection techniques identify a single set of features that is unambiguously influential on user-perceived stream quality, but that ultimately a combination of retransmitted and/or lost application-layer packets is most accurate for predicting stream quality.

An optimal service ordering for a world wide web server

We consider alternative service policies in a web server with impatient users. User-perceived performance is modeled as an exponentially decaying function of the users waiting time, reflecting the probability that the user aborts the download before the page is completely received. The web server is modeled as a single server queue, with Poisson arrivals and exponentially distributed file lengths. The server objective is to maximize average revenue per unit time, where each user is assumed to pay a reward proportional to the perceived performance. When file lengths are i.i.d., we prove that the optimal service policy is greedy, namely that the server should choose the job with the highest potential reward. However, when file lengths are independently drawn from a set of exponential distributions, we show the optimal policy need not be greedy; in fact, processor sharing policies sometimes outperform the best greedy policy in this case.

Revisiting a QoE Assessment Architecture Six Years Later: Lessons Learned and Remaining Challenges

In 2003, we presented an architecture for a streaming video quality assessment system [1]. Six years later, many of the challenges outlined in that paper remain. This paper revisits the 2003 architecture, updates it given what we have learned in our experience thus far with developing the architecture, and discusses in detail the remaining challenges to the realization of this architecture. We conclude with suggestions for moving beyond the biggest challenges, namely cooperation among the interested parties and system scale.

Systems considerations in real time video QoE assessment

We consider several key questions in the design of a real time video quality assessment system. How frequently can we generate subjective video quality ratings with some degree of accuracy? How often should we sample the data? How do we weigh the need to consolidate data collection (arguing for fewer, less frequent data points) with the need to monitor video quality in real time (arguing for more frequent data points)? What are the timing requirements for such a system, both in training the system and in assigning ratings to videos? Our results demonstrate that we can achieve accurate video quality ratings by using small portions of the video and a frequent sampling rate, and that in the worst case the system can be trained in ten minutes.

Computer science faculty careers at liberal arts colleges

DOES CS REALLY FIT IN A LIBERAL ARTS SETTING? Computer science both strengthens and is strengthened by the context of liberal education [1]. Computer science degrees at liberal arts colleges often emphasize multiple approaches I f you are a graduate student who loves teaching as much as (or even more than) you love research, you should consider a career in the liberal arts. As a prospective faculty member in computer science, you may be unaware of, or have misconceptions about, faculty careers at liberal arts colleges. Although we have all found tremendous satisfaction in our careers at highly ranked liberal arts colleges, most doctoral programs are undertaken at large research universities because many graduate students—and their advisors—are unaware of the career possibilities at liberal arts colleges. By sharing our collective experiences, we aim to dispel myths about teaching and research in computer science at liberal arts colleges. We want to give you a realistic basis for beginning to ponder the question, “Would I be happy and successful at a liberal arts college?”

A framework for self-healing home networks

Self-healing networks, or computer networks that can detect existing or potential pathologies and mitigate them with minimal human intervention, are particularly attractive in the home networking space, as home networks are heterogeneous and are typically configured and maintained by non-experts. Home networks greatly benefit from the ability to independently detect and mitigate issues with minimal user intervention. In this work in progress paper, we propose a proactive framework for a self-healing home network that detects and mitigates network pathologies that may lead to reduced application QoE. The framework collects and analyzes both application-level and network-level data to assessing the current “health” of the home network. In addition, the framework will apply a set of heuristics to determine the best course of action to take when presented with a set of network conditions.