Patrick Brown

Analysis of peer-to-peer traffic on ADSL

Peer-to-Peer (P2P) applications now generate the majority of Internet traffic, particularly for users on ADSL because of flatrate tarification. In this study, we focus on four popular P2P systems to characterize the utilization, the performance and the evolution of P2P traffic in general. We observe and compare the influence of each P2P application over the traffic, and we evaluate the evolution of these P2P systems over a year. Our analysis is based on ADSL traffic captured at TCP level on a Broadband Access Server comprising thousands of users. Thus, we characterize the P2P traffic and users, and we draw interesting results on connectivity and cooperation between peers, localization of sources, termination of connections and performance limitations. The evolution of the traffic over the year allows us to see the dynamics of the use of P2P systems. The difference between week days and week-end days informs us about the behavior of P2P users.

Resource sharing of TCP connections with different round trip times

The performance of TCP connections sharing a common link and the resulting link usage depends on the various round trip times along the connections. Expressions exist for these values only in the homogenous case and heuristics for the general case give uncertain results. In this paper we first derive an exact expression (in the setting of fluid models) for the window evolution of TCP connections with different round trip delays. We next apply this result in a particular context to derive expressions for the maximum window sizes and the throughput of TCP connections as well the idle periods and the utilization of the shared link. We show that a bias in favor of connections with short round trip delays persists even for large buffer sizes. First- and second-order approximations in buffer size are given respectively for small and large buffers. These expressions show that for small buffers small round trip times predominate in global performance while the opposite is true for large buffers.

Fairness analysis of TCP/IP

Bandwidth sharing between multiple TCP connections has been studied under the assumption that the windows of the different connections vary in a synchronized manner. This synchronization is a main result of the deployment of drop tail buffers in network routers. The deployment of active queue management techniques such as random early detection (RED) will alleviate this problem of synchronization. We develop in this paper a mathematical model to study how the bottleneck bandwidth will be shared if TCP windows are not synchronized. This permits to evaluate the improvement in fairness and utilization brought by the deployment of active buffers. Also, this indicates how much a synchronization-based study underestimates the performance of TCP in a nonsynchronized environment.

Detailed analysis of eDonkey transfers on ADSL

The large use of P2P systems gives rise to the question of the performance of these applications. The performance evaluation of P2P systems can be undertaken in many ways. Here, we focus mainly on file retrieval performance, i.e., the time to download a file, the throughput of downloads and the waiting time before the transfer begins. Our study is based on an ADSL concentration point with a large number of residential users. We analyze the TCP connections of eDonkey peers. Our measurement tool gives us the possibility to put together all the connections downloading the same file and thus, to study the general efficiency of file download. Our results indicate that the waiting time is important and independent of file popularity and that the average throughput is very low and has an unexpected dependence on file popularity. The performances vary significantly, such that on the one hand most of the transfers encounter low throughputs, and on the other hand some achieve very fast download rates

Batch Arrival Processor-Sharing with Application to Multi-Level Processor-Sharing Scheduling

We analyze a Processor-Sharing queue with Batch arrivals. Our analysis is based on the integral equation derived by Kleinrock, Muntz and Rodemich. Using the contraction mapping principle, we demonstrate the existence and uniqueness of a solution to the integral equation. Then we provide asymptotical analysis as well as tight bounds for the expected response time conditioned on the service time. In particular, the asymptotics for large service times depends only on the first moment of the service time distribution and on the first two moments of the batch size distribution. That is, similarly to the Processor-Sharing queue with single arrivals, in the Processor-Sharing queue with batch arrivals the expected conditional response time is finite even when the service time distribution has infinite second moment. Finally, we show how the present results can be applied to the Multi-Level Processor-Sharing scheduling.

Fluid Model for eMule File Sharing System

We propose a fluid model for eMule-like file sharing systems. The analytical analysis of a two file system is presented followed by the numerical analysis of a n-file system. We show the influence of the parameters such as the storage probability of a file and the request rate for a file on the system performance, i.e., on average downloading times. Our results reveal that the uploading policy, which equally treats each request, is not appropriate and that it introduces unfairness in the system. Afterwards, we discuss the optimal uploading policy on peers, i.e., which part of the uploading bandwidth should be dedicated to each shared file on a peer in order to improve the fairness in the system

Comparing FB and PS scheduling policies

In this paper we obtain new results concerning the expected response time of the foreground-background (FB) scheduling discipline and its comparison with processor sharing (PS). Some results previously derived for job sizes with finite second moment or bounded sizes, are extended to infinite second moments. New bounds and asymptotic results are also derived. We show that for job sizes with infinite second moment large jobs may benefit from the FB scheduling discipline although this discipline favors short jobs. For certain distributions all jobs sizes may even benefit from FB with respect to PS showing that the performance benefits obtained by some job sizes need not be obtained at the expense of others.

Large Scale Analysis of the eDonkey P2P File Sharing System

This paper presents a general numerical method to evaluate the download times in a P2P file sharing application with linear time and memory requirements with respect to the number of different files shared in the system. As this number typically exceeds several million and as the dependencies between file download times grow exponentially with the number of files, this method is useful to study realistic size systems. We apply the method on a system of 20000 files using parameters observed on a real network. The resulting performance is discussed. We show how the method may be extended to take into account initial waiting times before download and present the resulting performance.

Optimal choice of threshold in Two Level Processor Sharing

We analyze the Two Level Processor Sharing (TLPS) scheduling discipline with the hyper-exponential job size distribution and with the Poisson arrival process. TLPS is a convenient model to study the benefit of the file size based differentiation in TCP/IP networks. In the case of the hyper-exponential job size distribution with two phases, we find a closed form analytic expression for the expected sojourn time and an approximation for the optimal value of the threshold that minimizes the expected sojourn time. In the case of the hyper-exponential job size distribution with more than two phases, we derive a tight upper bound for the expected sojourn time conditioned on the job size. We show that when the variance of the job size distribution increases, the gain in system performance increases and the sensitivity to the choice of the threshold near its optimal value decreases.

Unfairness in the e-mule file sharing system

Measurement studies have shown that the probabilities of file possession and the arrival rates of file requests have an important impact on the file downloading times in p2p file sharing systems resulting in unfair performances. A model has been proposed to capture this phenomenon. This model is used to derive the qualitative impact of system parameters on performances. However to obtain results for realistic p2p networks one is confronted with the large number of files in these systems. To show the influence of the number of files shared in the system we present results obtained by simulations in the case of small to medium size systems.