From: Yu Hu I read the two papers, the first one gives me an mathematic and statistic impression whileas the second one gives me just a plain history report. At the sameI can't figure out the tight relationship between the first and second. However, anyway, I champion the first paper and accept plainly the second one. First: Best-Effort versus Reservations [ Lee Breslau and Scott Shenker ] Best-Effort-Only architecture and Reservation-capable architecture are two choices discussed here. The paper main idea is comparing them in terms of performace and total welfare. (The two architecture should be introduced more, especially Reservation-capable , I want to make it clear whether the reservation-capable's reservation function or whatever are implemented in the core of network, which will be related the class 1 papers "end-to-end argument"? ) Since network isamazingly complicated, we must measure it in some model, such as Fixed Load Model and Variable Load model. In Varable Load model, the paper analyzed six cases(three load distributions times two kinds( application kind : rigid & adaptive)). Based on analyze of a cluster of results, adding results of extension to model, authors got their conclusion as comparison between the two architectures highly depends on load patterns. Anyway, the selection is hard to make since manyaspects of future network developing so rapid and diversely. I champion the paper not just because it analyze and compare two architectures using models, but also it help me to learn to analyze something in mathmatical and well-defined way. I think it is a good paper about internet permance and cost analysis and comparison. Second : Internet pricing and the history of communications [Andrew Odlyzko] The paper is not a techinical paper, its key idea is straight forward: people are willing to pay more for simple pricing plans. I admire author for that he can find so a lot of history evidence and corresponding statistical data. The paper mentioned some somple plan such as flat-rate which continues to dominate thedata transmission market.In spite that there exit some couterexamples, the general conclusion is hold: strive for simplicity, even at the cost of efficiency. The point of the paper is pretty trivial and not worthy of researching , but worthy for merchants. I don't think it is a good paper, it is too plain. After reading it ,I just say " ok, yes, so ..." . Anyway, it really has no much merits except its length. YU HU ############################################################################# Date: Thu, 4 Apr 2002 16:21:04 -0600 (CST) From: Xinghua Shi To: Adriana Iamnitchi Subject: Summary2 Reading2: Paper1: Best-Effort vs. Reservations. Contribution: This paper provides several observations that supports best-effort only architecture or reservation-capable one, using a simple analytical model. First, it presents the simplest model C Fixed Load Model. Because in reality the applications are elastic, this model cant assure us that the extent of the performance advantage of the reservation capable architecture. Then, the authors propose a variable load model of discrete and continuum version. From analysis of the probable results of the model, the paper comes out the discussion of the tradeoff between reservation-capable and best-effort-only architectures. This gives us a good example in analyzing performance or comparing different architectures by building simple mathematic models and doing some quantity analysis. Disadvantages: This paper doesn't point out clearly which architecture is best though it gives analysis on these architectures based on some mathematical models. There are so many factors to consider in selecting an architecture in reality which can't summarize with simple models. It also doesn't show to what extent the models approximate the real performance of architectures, so the results aren't so powerful in the practical use. Questions: This paper uses different approximate functions to plot the behavior for high bandwidth and low bandwidth(b). Are these approximations converge since this function maybe the same indeed? Paper2: Internet pricing and the history of communications Contribution: This paper suggests that the Internet will evolve towards simplicity as other previous communication technologies from historical analogies. It's a practical analysis because it considers economic,political and social issues . It also covers a general analysis using quantitative and qualitative measures in the communication technology trend. Besides, due to the unique characters of the Internet, this paper discusses some factors that might make historical analogies inbalid. Disadvantages: As we know, the Internet is a quickly changing world. So there are many new applications and technologies added day by day. Then maybe all these ananlyses based on the historical views of communication world will be incorrect because of some breakthrough technology coming up in the Internet. Questions: As long as the development of the Internet, is flat rate the paper favors still be feasible and preferable? Can we predict some new applications or technologies in the Internet which will defy the proposals in the paper? Best Mindy ############################################################################# Date: Thu, 4 Apr 2002 18:34:44 -0600 (CST) From: Rahul Santhanam Breslau-Shenker: 1. The main contribution of the paper: It analyses the cost/utility tradeoffs between best-effort-only architectures and reservation-capable architectures for a variety of network load distributions. 2. a. 4 - significant contribution. The paper provides a balanced and rigorous perspective on the questions it addresses. b. The results in the paper seem correct. Also, I have no problems with the terms in which the debate is formulated. The network load distributions considered were probably chosen because calculations with these distributions support differing perspectives on the viability of reservation-capable architectures as opposed to best-effort-only architectures. Unfortunately, the question of whether these distributions are plausible is not addressed in sufficient detail (The authors could have provided some empirical data on the nature of current load distributions, which would presumably not be uncorrelated with load distributions in the future, even if the latter are hard to predict.) More importantly, the authors do not sufficiently justify the form of the utility function they consider. Perhaps they could have considered a variety of utility functions (at the cost of making the paper more complex..)? c. The most important limitation of the approach is that, as the authors themselves put it, "our results are frustratingly ambiguous on the fundamental question of which architecture is best". 3. Strengths of the paper: 1. Clarity in presenting the argument 2. Mathematical rigor of the approach 3. Reasonableness of claims made 4. Weaknesses of the paper: 1. Too much detail. It's not clear how applicable the results are anyway, so why analyse them so closely? 2. Naive modeling of the economic tradeoff. It is unclear how to model the cost of installing one architecture as opposed to another - an indication that the discussion is taking place on too abstract a level. 5. Question I would like to ask the authors: How does your stated preference for reservations square with your results showing that for both Poisson and exponential load ditributions, best-effort-only architectures are likely to be the best choice? The results do show that reservation-capable architectures are preferable for algebraic distributions. But (1) the evidence presented in this paper that netwrok loads might be described by algebraic distributions is very weak indeed (2) the results are alarmingly unstable with respect to fine-tuning of the model for algebraic distributions, unlike in the case of Poisson and exponential distributions, so the validity of the results with respect to a "perfect model" is open to question. 6. Extensions of the work: To do the analysis with a load distribution clealry supported by empirical data. Perhaps analysis of how reservations are made could be done based on the parameters of a flow, i.e., different reservation-based architectures could be compared? QoS? Rahul. ############################################################################# Date: Thu, 4 Apr 2002 23:57:33 -0600 (CST) From: Catalin Lucian Dumitrescu Best Effort versus Reservation: A Simple Comparative Analysis ------------------------------------------------------------- 1. State the main contribution The paper presents the comparative analysis of best-effort vs. reservation-capable in the transmission of network packets for different cases. Best effort consists in packet delivery without guarantees or permission asking. In contrast, reservation-capable approach introduces bandwith-reservation abilities and admission control over packets. A simple model is initially introduced, the fixed load model, and extended later on several dimensions. This model assumes that over a link of capacity C, k network-flows exist. The bandwith is equally distributed (C/k). In addition, flows are associated with applications whose perfomances are the same function pi dependent on the individual bandwith b - pi(b). In addition to individual utility function, the network has also associated an utility function V(k) = k*pi(C/k). Function of the V(k) landscape (modeled by the specific hardware resources), one or another approach becomes appropiate. The analysis focuses on several dimensions: application behavior (elastic, rigid), load model (fixed, variable), and capacity properties (fixed, variable). Two additional extensions are presented in the end, sampling and retrying. 2. Critique the main contribution The analysis does not take in account: - underlying implementation issues and models (where and how) - assumes a very simple network (one link) with static routing - in fact the focus is on packet transmission between two points and not on a more complex system that can deploy dynamic routing. It is on network utility and not on the application gains. It measures the amortization cost of the network, not the utility of individual applications (not end to end) a. Rate signifiance - 3 b. Rate the methodology - model is to simple, no experimental part is involved. Some experiments may cast some light and also to push analysis in unexpected directions that may prove more useful in real cases 3. Three strongest points - model simplicity - clear analysis in several specific directions - completeness of analysis 4. Three striking weaknesses - a lot of math and formulas - no validation of the theoretical results (simulator - simulations) - avoidance of a clear conclusion (not just formulas) 5. Three questions for the author - none 6. Detail an interesting future work - do some simulations and compare resuts - some additional factors may change results 7. Optional comments - none ------------------------------------------------------------------------------- Internet Pricing and the history of communications ------------------------------------------------------------- 1. State the main contribution A long survey of about pricing history of different technologies, maybe to draw some conclusions to be used for Internet also. "History of communication suggests strongly that as services become less expensive and are more widely used, the balance shifts away from the need to segment the market, and thereby to extract maximal revenues and to maximize efficiency of the infrastructure" 2. Critique the main contribution a. Rate signifiance - b. Rate the methodology - 3. Three strongest points - 4. Three striking weaknesses - 5. Three questions for the author - 6. Detail an interesting future work - to wait several years (or decades) and see if the Internet is different than telephony 7. Optional comments ##################################################################### Date: Fri, 5 Apr 2002 10:10:10 -0600 (CST) From: matei@cs.uchicago.edu Lee Breslau and Scott Shenker, Best-Effort versus Reservations: A Simple Comparative Analysis, SIGCOMM98 1. The question this paper helps answering is: should the Internet keep its best-effort architecture or should it move towards a reservations capable framework. 2a. Rate 4. A large number of projects propose technical solutions for best effort or reservation capable networks. This paper brings a quantitative framework to compare benefits and costs of these two approaches. 2b. Rate 5. I found this paper compelling providing a good framework for duscussion and a roadmap for future refinements. Of course the utility and cost functions used are debatable but the authors are careful to say their examples provide only intuition. 3. Strong points: a pretty general model. It moves the focus on choosing the 'right' utility and cost functions, and on forecasting usage patterns. 4. There are a number of apparently simple questions that could have been answered: - what's (the empirical evidence for) number of flows distribution in the present network. What are current trends? What happens in current telephone networks? - it's quite likely that the cost for an adding a unit of bandwidth is not constant (as the paper mentions) but decreasing. The more bandwidth we add the incremental cost per unit is smaller. This will change the model in Sec. 4. - it is likely that reservations and best-effort schemes will coexist on the same networking infrastructure - so the costs for better network provisioning are probably smaller than those presented in the paper. 5. Questions to authors: - what would be an adequate pricing scheme for a network that provides both reservations and best-effort transport. - how did traditional telecom decide their levels of provisioning? What were the patterns in these networks? Andrew Odlyzko, Internet pricing and the history of communications, Computer Networks, 2001 1. The paper presents a number of historical examples of communication services pricing. Most examples are geared to support the idea that, in a consumer market, as communication services become less expensive, consumers prefer the simplest pricing scheme. 2a. Rating 3. It seems to me that although this collection of examples and empirical evidence is useful and entertaining it does not significantly push the flied further. 2b,c. Rating 3. Merely collecting facts is unlikely to support claims of generality. On the other side I do not have a better idea on how to improve this. 3. Consumers prefer simple pricing mechanisms even though these pricing schemes do not reflect (sometime lower) usage volumes. As a result, higher level services (reservations, QoS, etc) might not be accepted by consumers since these services generally require more complicated pricing schemes. The author however does not discuss the fact that markets for these services might develop separately from the general Internet providers market (e.g. there might be a market for video-calls completely separated, and with different providers). 4. I'd have liked to see some modeling to answer questions like: How low has the cost of communication to decrease before providers loose their pricing power and consumers manage to impose their preferences for flat rates? How fast these processes are? 5. Questions: Large-scale multicast deployment on the IP infrastructure was largely unsuccessful. How does this relate with the currently used pricing mechanisms? How does one produce and validate a model for consumer preferences for pricing schemes? How service dependent is this. 6.