Gregory B. Brewster

Internet packet loss: measurement and implications for end-to-end QoS

We analyze a month of Internet packet loss statistics for speech transmission using three different sets of transmitter/receiver host pairs. Our results indicate that packet loss is highly bursty, with the majority of individual losses occurring in a relatively small number of bursts. We find that loss exhibits dependence in most cases, but is not always well modeled as dependent. We introduce an analytical technique for measuring loss dependency. We also consider the asymmetry of round trip packet loss, and find that most loss on a round trip path occurs in either one direction or the other. We introduce a normalized metric for measuring loss asymmetry and apply it to our measurements. Finally we discuss the implications of our study for the next generation of real time voice services in the Internet.

Analysis of Topology Aggregation techniques for QoS routing

We study and compare topology aggregation techniques used in QoS routing. Topology Aggregation (TA) is defined as a set of techniques that abstract or summarize the state information about the network topology to be exchanged, processed, and maintained by network nodes for routing purposes. Due to scalability, aggregation techniques have been an integral part of some routing protocols. However, TA has not been studied extensively except in a rather limited context. With the continuing growth of the Internet, scalability issues of QoS routing have been gaining importance. Therefore, we survey the current TA techniques, provide methodology to classify, evaluate, and compare their complexities and efficiencies.

Self-similarity of Internet packet delay

In this paper we present results which suggest that Internet packet delay, when viewed as a time series, is self-similar in nature. A self-similar phenomenon displays the same or similar statistical properties across a wide range of time scales. Self-similarity has previously been observed in the magnitude of traffic transmitted on local-area and wide-area networks. Our research is the first to explore the self-similarity of traffic delay rather than traffic magnitude. Our results imply that the delay experienced by application layer protocols is extremely bursty and that traditional Poisson models cannot accurately predict the degree of burstiness. Furthermore, we present evidence that the degree of self-similarity for a round-trip path in the Internet may be correlated with the packet loss observed on that path.

Measurement and analysis of long-range dependent behavior of Internet packet delay

We analyze 12 traces of round-trip Internet packet delay. We find that these traces, when viewed as time series data, often exhibit Hurst parameter (H) estimates greater than 0.5, indicating long-range dependence. Several traces, however, are not well-modeled with a constant H. We discuss in detail our analytical methods and the robustness of empirical estimators of H under conditions of non-negligible packet loss. Our results indicate that Internet delay is bursty across multiple time scales, which implies that end-user duality of service in the Internet is likely to be impacted by long periods of very large and/or highly variable delays.

On the impact of loss and delay variation on Internet packet audio transmission

The quality of audio in IP telephony is significantly influenced by various factors, including type of encoder, delay, delay variation, rate and distribution of packet loss, and type of error concealment. Hence, the performance of IP telephony systems is highly dependent on understanding the contribution of these factors to audio quality, and their impact on adaptive transport mechanisms such as error and buffer control. We conducted a large-scale audio transmission experiment over the Internet in a 12-month-period in order to evaluate the effects and the correlation of such parameters on audio transmission over IP. We have noticed that the correlation of loss and delay is not linear, but stronger correlation is observed as the delay approaches certain thresholds. We have made a number of new observations on various delay thresholds that are significant for loss prediction for adaptive audio transmission over IP networks. We also have made new observations to assess the audio quality of PCM @m-law and G.728 codecs under different loss and delay conditions. The paper provides a number of recommendations for implementing efficient adaptive FEC mechanisms based on our measurement observations and analysis.

Audio transmission over the Internet: experiments and observations

The performance of IP telephony systems is highly dependent on the audio codecs and their reaction to packet loss and instantaneous delays. Understanding the interaction between audio encoding and the dynamic behavior of the Internet is significant for designing adaptive audio transport mechanisms. For this purpose, we conducted a large-scale audio transmission experiment over the Internet in a 12-month period using various Internet sites. As a result of this experiment, we have made a number of new observations to assess the audio quality of G.711 and G.728 codes under different loss and delay conditions. The paper also states a number of recommendations for implementing efficient adaptive FEC and playout mechanisms.

The fairness of DQDB networks with slot reuse

The paper analyzes the fairness characteristics of two previously proposed slot reuse protocols for DQDB networks. In particular, the slot reuse protocol proposed for the IEEE 802.6e standard [Hassanein et al., 1994] results in unfair steady state bandwidth allocations for some overloaded DQDB networks and can cause unbounded access delays under some stable workloads. In contrast, a previously proposed simple counter protocol results in bandwidth allocations in overload that are shown to be at least weakly fair for all workloads and optimally fair for many important workloads. This simple protocol also guarantees bounded access delays for any stable workload. A new modified counter protocol is proposed that has superior fairness characteristics to either of the previous protocols.

On studying the impact of the Internet delays on audio transmission

The quality of the audio in IP telephony is significantly influenced by various factors, such as type of encoder, distance, delay variation, rate and distribution of packet loss, type of error concealment, and others. Hence, the performance of any IP telephony system is highly dependent on understanding the contribution of these factors to audio quality, and their impact on adaptive transport mechanisms such as error and buffer control. We conducted a large-scale audio transmission experiment over the Internet in a 12-month period in order to evaluate the effects and the correlation of such parameters on audio transmission over IP. As a part of studying and analyzing the collected data, we have made a number of new observations on the correlation of loss and RTT (round trip time) variation, and various RTT measurement mechanisms that are significant for adaptive audio transmission over IP networks.

Empirical studies and queuing modeling of denial of service attacks against 802.11 WLANs

The growing popularity of 802.11-based wireless LANs (WLAN) also increases the risk of security attacks. Most studies of WLAN security are on the protection of data integrity, and few studies are addressing the issue of Denial of Service (DoS) attacks. This paper studies two major DoS attacks of authentication request flooding (AuthRF) and association request flooding (AssRF). Our studies show that these DoS attacks cause significant performance degradations and may disconnect the communications. A queuing model is presented to study the attacking mechanisms, and the causes of performance degradations. The analytical results of the queuing model are validated by the simulation model, and both results are consistent with the empirical data. The queuing model analysis leads to the development of four solutions: Request Authentication (RA), Reduction of Duplicate Requests (RDR), Reduction of Response Retransmissions (RRR), and Round Robin Transmission (RRT). We tested these four solutions and collected empirical data to validate the effectiveness of the solutions. A comparison of these four solutions is presented to show their strengths and weaknesses in resolving the attacks.

Performance analysis of application response measurement (ARM) version 2.0 measurement agent software implementations

Effective distributed application performance management infrastructures are based upon the collection and management of key performance metrics including resource utilization, workload and service measurements. Without application oriented transaction service measurements, the viability and long term usability of the client/server paradigm for mission critical applications is in question. The Application Response Measurement (ARM) Application Programming Interface (API) addresses this requirement by enabling the measurement of transaction level metrics, notably response time, for distributed applications through application code instrumentation. Version 2.0 of the ARM API includes the ability to measure and correlate response time components on different processors (i.e., clients and sewers). ARM Agents encompass the software components responsible for interception and processing of the ARM API calls including support for response time measurement, collection and communication. This paper introduces alternatives for several important design aspects of ARM agent software that are not specified in the ARM standards documents. In particular, we consider the architecture of ARM Agent software, correlator generation and communication, and ARM data storage. We propose, implement and analyze the performance of several alternative designs. This work provides an assessment of ARM instrumentation impact on application response times as well as insights into the design issues involved. We first provide a general overview of the ARM 2.0 API and features. We then consider alternative ARM agent software designs. Next, we present measurements results for several design implementations that shed light on the overhead involved with instrumentation and the developed designs. Finally, we present conclusions and a description of future work.