Date: Tue, 9 Apr 2002 16:52:44 -0500 (CDT) From: Yu Hu To: Adriana Iamnitchi Subject: Preview of class 3 (YU HU) On the Marginal Utility of Network Topology Measurements [Paul Barford, Azer Bestavros, John Byers, Mark Crovella] Now, for the future of internet, analyzing it strcture and behavior is of some significance, needing us to do some deep and correct measurements.In this paper,author focussed on their further work on the Marginal utility of network topology measurements.Traceroute, which is one of approach of Network topology measurements, is used here for internet topology discovery. Authors assumed internetas a model made up of leaf nodes,boders,backbone,stub.Authors showed us a very theoretical view and experiments' results, convincing me much about the internet topology. The contribution of the paper, I think, is producing us a manageable model to measure internet topology using basic internet components such as links , nodes and nodes' degree. Of course, if you add more measurements, it willprovide a more complete picture of the entire topology. Some drawbacks of the model that the paper showed to us, are the topology was sensitive to the particular choice of measurement sites and belief that further measurements were warrented to reinforce the model. All these leave open to attack on the conclusion of the model and future optimum. The most impressed me of the paper is that it freshly showed us a quantive view and analysis on the marginal utility of network measurements, which will surely lead to more work on it. So I champion the paper. On the Constancy of Internet Path Properties [Yin Zhang, Nick Duffield,Vern Paxson, Scott Shenker] Unlike other papers concerning on how measurements retrieved the charateristicsof internet and guided the design of internet, this paper focussed on the constancy of these measurements. Authors gathered two basic types of measurements toanalyze. Mathematical constancy,operational constancy,predictive constancy are three different ,but overlapping somewhat,notions of constancy.In the former two basic measurements,the paper showed us a lot of work about the three constancy. The main contribution of the paper, I think, is that it showed us a model and a method to analyze the constancy charater of these mearsurements. The paperhas shed light on the current degree of constancy of loss, delay and throughout,whileas got some interesting findings that leading us to more work on the model. Anyway, to make sure how steady the internet is need many aspects of constancy and datasets. The paper is also fresh to me , and "steady" is interesting and of significance to the measurement. So I champion the paper. #################################################################### Date: Tue, 9 Apr 2002 17:59:59 -0500 (CDT) From: Xinghua Shi To: Adriana Iamnitchi Subject: Reading3 summary Hey, This is my summary for tomorrow's papers. Reading3: Paper1: On the Marginal Utility of Network Topology Measurements. Rate:3.5 Contribution: The authors discuss some issues related to the network discovery problems to show how this study helps clarify how end-to-end paths pass through the Internet. They also study margunal utility from a purely empirical perspective and a theoretical perspective as well. They provide detailed results with plots, along with some ananylises based on this.In all, they come up with some insight into the Internet measurement, which deepens our understanding of the Internet behavior somehow. Disadvantages: The research they proceed are based on some assumptions which may be incorrect with the development of the Internet. Questions: How to chacterize those regions of the Internet topology which are relatively difficult to uncover using traceroute? How much effect does this characterization have on the current reseach work? Paper2: On the Constancy of Internet Path Properties. Rate:4 Contribution: The authors investigate three notations of constancy in the Internet measurement (Mathematical, Operational, Predictive)by measuring three quantities describing Internet paths(Packet loss, packet delay and throughput). They measured a large volume of Poisson packet to assess loss and delay characteristices which makes their discission somewhat sound from emperical view. They also use some new approach to study the above issue. Their attempt to find some fundamental concepts and tools for Internet measurement. This idea is usable and applicable in the quicky-changing Internet world. Questions: Are the concepts and methods this paper presents are useful indeed as more applicatons and protocols in the Internet emerge gradually? How to use these ideas to guide the design of the future Internet applications or organization? Thanks! Best Mindy #################################################################### From: "Catalin Dumitrescu" To: Subject: review Date: Tue, 9 Apr 2002 22:54:22 -0500 On the Marginal Utility of Network Utility Measurements ------------------------------------------------------------- 1. State the main contribution (or focus?) The paper addresses an approach to "network tomography" required by the Internet structure and its current technical approach. It is focused on = providing insight about using "projections", in which arrays of different sources = and destinations are considered in order to discover nodes, links, degree of backbone nodes distribution. Even more, the main concern is the analysis = of "marginal utility", the analysis of "the more is better" approach. 2. Critique the main contribution Even the theoretical results are accurate, experimental results and additional technical "vague" issues may introduce unexpected deviations of results. I think that most of these are also presented in the "Experimental Methodology" chapter. a. Rate significance - 2 b. Rate the methodology=20 - 4 3. Three strongest points - theoretical support - experimental data and additional analyses - final section with additional theoretical analysis of marginal utility - node disambiguation procedure -------------------------------------------------------------------------= ------ On the Constancy of Internet Path Properties ------------------------------------------------------------- 1. State the main contribution (or focus?) In this paper authors explore different constancy areas (mathematical, operative and predictive). Operative constancy is largely used as a = means for improving quality of services by different applications and = protocols. One dimension of the study tried to shed light on three Internet = properties: packet losses, packet delays, and throughputs. On another dimension, the = paper comes with identification of elements in the three areas of = constancy. 2. Critique the main contribution a. Rate significance - 2 b. Rate the methodology=20 - 3 3. Three strongest points - used data (lot of relevant data) - involved mathematic elements (fine analysis) - classification of measurements 5. Three questions for the author - how reliable is zing? (especially after the note in page 3) 6. Detail an interesting future work=20 - apply the same strategy to Gnutella 7. Interesting work - do the same experiments for each continent, and for all #################################################################### Date: Wed, 10 Apr 2002 00:54:05 -0500 (CDT) From: Ivona Bezakova To: Anda Iamnitchi Subject: class 3 - papers On the Marginal Utility of Network Topology Measurements -------------------------------------------------------- 1. The authors analyze Internet topology using "traceroute" method: run traceroute from every source (8 nodes) to every destination (1277 nodes) and then merge the resulting graphs. - Contribution 1: Authors show that at least 15% of the nodes use more than one IP address - multiple copies have to be merged appropriately (disambiguation procedure). - Contribution 2: Authors claim that as the number of sources increases, the likelihood of discovering a new vertex decreases significantly - experimental evidence and theoretically well-defined measure of success ("coverage") are provided. - Contribution 3: Authors define on-line marginal utility as the average information gain (expected difference of the original and the next entropy function after adding a new source). Interesting statistics about on-line and off-line marginal utility for additional sources and destinations are shown but no conclusions are drawn. 2. a) The paper is significant contribution (as far as I can judge it) because it defines new measures that could be used to compare their results with results of similar experiments in other publications. b) The paper is well-written and methodology is convincing. 3. Strongest points: Definitions of theoretical models underlying the experiments and experimental support for them. 4. Weakest points (and extensions): Distribution of source nodes. Sources from non-research environment may exhibit different behavior. Comparisons to previous papers w.r.t. the new measures. On the Constancy of Internet Path Properties -------------------------------------------- 1. The authors study three types of constancy of Internet paths: mathematical (single time-invariant mathematical model, e.g. some parameters don't change for the given dataset), operational (parameters remain within bound considered operationally equivalent), and predictive (predicting new characteristics using a window of past measurements). Constancy is experimentally measured for three functions of Internet paths: loss, delay, and throughput. Results: In general, coarser time scales exhibit more constancy than the finer ones. Loss happens in "bursts", suggesting constancy behavior - 77-85% of the regions (defined from the start of a loss episode until the next loss episode) are mathematically steady (to be modeled by exponential distribution, loss episodes by Poison). Mathematical and operational constancy mostly coincide. Predictive constancy (not depending on the choice of predictor) is worse because the loss episodes behave as independent identically distributed random variables. Packet delays are not mathematically neither operationally steady, but they are highly predictable. For throughput, some mathematical constancy can be observed, for operational constancy an additional parameter has to be specified - leading to exhaustive statistics, not corresponding to the mathematical constancy. On the other hand, throughput is well-predictable if the time scale is fine enough (minutes, not hours). 2. a) The contribution of the paper is significant because the notion of constancy seems to be an important property assumed by many practical algorithms (probably without any underlying research or statistics). b) The methodology is convincing although I do not know the statistical tests used in the paper, neither the NIMI measurement infrastructure, so I was not able to verify soundness of most of the techniques and claims. #################################################################### Kavitha: On the Marginal Utility of Network Topology Measurements 1. The paper measures the marginal utility of adding addition measurement sites , used to map the structure of the internet. The marginal utility declines rapidly after adding a second or third site. 2. Contribution- 3 THe first paper to quantify utility of adding measurement sites. The methodology was pretty convincing as they had both emperical evidence and a theoretical analysis. 3.Interesting parts: That the internet topology is not a random graph but has a distinct core and feeder links. The methodology to disambiguate router interfaces by sending UDP packets to unused ports. 4. Found it difficult to stay interested in the extended results. All in all a refreshing paper which tries to capture the mystery of the internet topology. h