Date: Thu, 02 May 2002 21:29:01 -0500 From: Oleg Pashko Resilient Overlay Networks 1. RON enables distributed Internet applications to detect and quickly recover from path outrages and periods of degraded performances. It allows a group of nodes to communicate with each other when problems with underlying connecting Internet paths occur. 2a.3.5.Practical addition (at least in some settings of) to the existing Internet routing scheme. 2b.Simulations seem to be well designed and network experiments show good results. 2c.Possible bandwidth waste as a result of aggressive path probing by nodes. Expensive overlay maintenance protocol.Limited settings of RON. Potential presence of several RONs competing over Internet paths. 3.Choice of a routing metric - 1) latency, 2) packet loss rate, 3) throughput Quick path recovery times of RON (seconds). 4.same as 2c. + RON performance in p2p settings? 5-7. Have other possible metrics ever been tried? Would it be better to keep some rating of a virtual link as seen by nodes.How well it can work in p2p systems. Exploring Network Proximity in p2p Overlay Networks 1.A study of the network locality properties of a p2p overlay network, Pastry. 2a.3. More refined overlay network (in the spirit of CAN, Chord & Tapestry) is proposed. 2.b. Simulations are OK, yet, perhaps more (than 2) different kind of Internet topologies should be considered. Experiments with real p2p systems are also desired. 2.c.More complicated protocol (than in Chord, say). that adds to the overall network complexity. Performance analysis of applications build on top of Pastry. 3.Locality heuristic that allows to exploit the network proximity incorporated into the Pastry platform is a plus. Node join an failure analysis with the improved protocol are interesting. Paper itself is very detailed with a good mathematical foundation of the algorithm. 4.same as 2c. 5-7.How well the synthetic network topologies (on which simulations and experiments were run) capture real Internet structure? What is the acceptable overhead trade off to capture the proximity of the underlying Internet? Does the proximity invariant hold for more realistic Internet topologies ############################################# ############################################# Date: Thu, 2 May 2002 21:55:53 -0500 (CDT) From: Rahul Santhanam Castro-Druschel-Hu-Rowstron: 1. The paper is a study of how the PRP overlay network Pastry exploits network proximity. it incorporates analysis and simulations using network topology simulators. 2. a. 3 - modest contribution. b. The experiments are well-designed but the results are not well-presented. (e.g. For no good reason, the bars for Normal Routing and Perfect Routing are switched in the graph for per-hop distance vs bar number for GaTech topology as compared to Sphere topology). Some counter-intuitive results, such as that normal routing tables perform better than perfect routing tables for hop number 5 in the graph for GaTech topology, are not explained. Distance ratios for proximity-based protocols are not compared with distance ratios for protocols that do not use proximity. Also, the experiments address only some aspects of the costs of proximity-based protocols. c. The most important limitation of the approach is that it is specific to Pastry, unlike what the authors claim. 3. The idea of evaluating how well proximity information can be exploited is an important one. 4. (1) The paper is badly written. The design of the Pastry Network is not presented clearly, and the notation used in the theoretical analysis is cumbersome. There doesn't seem to be any point to the inclusion of the theoretical section. (2) The proximity-based protocols do not specify what is to be done with leaf sets. 6. Perhaps the process of determination of node ids could also use proximity information to increase efficiency? Andersen-Balakrishnan-Kaashoek-Morris: 1. The paper describes an architecture - "Resilient Overlay Network" - that allows distributed Internet application to detect and recover from Internet routing failures. 2. a. 3 - modest contribution. The architecture described is effective but does not have general application. b. The experiments are well-designed . The amount of data collected is huge - sufficient to validate the reuslts. The experimental results are clearly presented. The hosts could have been chosen to be more widely spread out over the Internet,though. 3. The results are impressive - RON is able to detect and recover from path outages in a few seconds. 4. The RON architecture is high-cost and special-purpose. 6. To construct scalable RONs, perhaps by decoupling them from applications. ############################################# ############################################# Date: Thu, 2 May 2002 23:08:16 -0500 (CDT) From: Xinghua Shi Paper1: Resilient Overlay Networks Contribution: The paper addresses a Resilient Overlay Network(RON) which is an architecture that equips distributed Internet applications with fault detection and recovery work. All the improvements of RON demonstrate the benefits of moving some of the control over routing into the side of end-systems. This is an example of the end-to-end argument we discussed before in the class. The authors first describe the design and implementation of RON and then present several experiments that evaluate and demonstrate the benefits of RON. Rate: a. 4 in significance: This paper has significant contribution because first wide-area network overlay system that can detect and recover from path outrages and periods of degraded performance within several seconds. a. Rate: 4 in convincing of methodology: The paper is well organized and has detailed technical explanations. The authors use a IP-layer protocol BGP-4 as a comparison in the argument which makes it convincing. They also provide three critiques of RON and thus makes a complete argument. The strongest and/or most interesting ideas in the paper: a. RON can route around most failures by using only one intermediate hop(RON is so effective?). b. Their current sixteen-node RON deployment that spans different places and diverse sites. c. They found that the benefits of quick fault detection and successful recovery are realized on the public Internet and do not depend on the existence of non-commercial or private networks. d. RON is the first wide-area network overlay system that can detect and recover from path outrages and periods of degraded performance within several seconds. e. Some historic background and comparisons before dealing with RON design and implemention details. Questions: a. Is it feasible to explore multi-criteria path selection? How about its cost vs. performance? b. How to decide which RON node is entry node and which is exit node? Are they pre-determined by the RON architecture? c. How about the cost and latency of RON performance database? Paper2: Exploring network proximity in peer-to-peer overlay networks Contribution: This paper provides a comprehensive study of Pastry's locality properties via analysis and via simulations based on two large-scale Internet topology models. They show that network locality properties hold also in more realistic network topologies based on simulations on two different Internet topology. Rate: a. 3 in significance: This paper has moderate contribution because it presents an anlysis which has realistic meaning in the exploring network proximity but there are not many novel ideas in the paper. b. 3 in convincing of methodology: The argument is based on the theretic analysis and experimental simulations. But the argument is merely based on experiental data and results on Pastry and lacks comparisions, which weakens the conclusions. The strongest and/or most interesting ideas in the paper: a. The detailed analysis of Pastry's locality properties. b. The formal definitions and thermos c. They build simulations on two different topologies. d. The algorithms allow newly joining nodes to efficiently discover a nearby node in the existing Pastry overlay. The most striking weaknesses in the paper: a. Their analysis and simulations are based on Pastry, but they claim many of their results apply to Tapestry and Chord without convincing proof. Questions: a. Their analysis and simulations are based on Pastry, so how can they claim many of their results apply to Tapestry and Chord? ################################################# ################################################### Date: Thu, 2 May 2002 23:44:21 -0500 (CDT) From: Yu Hu YU HU Resilient Overlay Networks [David Andersen,Hari Balakrishnan and etc] Main contributeion of the paper: This paper depicted to us a good platform: Resilient Overlay Network(RON).Author explained in theory why RON could detect and recover from outages and path failures more quickly than current inter-domain routing protocols, and introducedto us the design rule and operation mechanism of RON. After evaluating two datasets, authors gave the conclusion that RON is a good overlay network, from which a variety of resilient distributed Internet applications can be benefit.I thiink the main contribution is rational and introducion of RON. Critique the main contribution: Significance : 4( significant contribution) Methodology: Firstly, this is a introduction paper. From design goals, design issues to implemention, authors made us get to know RON step by step. After that, authors evaluated the performance of RON by some experiments with a wide-areaRON deployed at several internet sites. These methods and results coming from them are convincable to me. The most important limitation of the approach: The datasets used in experiments RON1 and RON2, maybe authors should give more details about them, maybe should tell readers the difference between the two datasets, can these datasets represent real network at least in some sense. Most interesting ideas: In intuition, when failures happened, the detecion and recoverment will put a lot of load on network. However, RON will accomplish the task at a low cost. Itsidea is interesting and worthy being researched. Weakness&Questions: 1. From RON1 and RON2's experiments results, can we conclude that RON is a goodplatform in real network ? 2. Is the RON's policy mechanism dyniamical ? How does it deal with user want to add some new policies or delete some old policies? 3. When you forward your message using RON, how do you deal with the security problem, such as: if you don't want to let some node to know your message, but RON forward to it? Comments: This is a good paper. It maybe give us a direction to deal with route failure and recover: move function to application layer, construct overlay(example RON)and route based on it sometime. Exploiting network proximity in peer-to-peer overlay [Miguel Castro, Peter Druschel,Y.Charlie Hu,Anton Rowstron] Main contributeion of the paper: Peer-to-peer overlay network's locality property is pretty useful for user application. This paper main contribution is that it gave a detailed and deep analysis on the property and using some experiments to evaluate them. The paper madeus trust that exploiting network proximity is practical at low cost. Critique the main contribution: Significance : 4( significant contribution) Methodology: Authors based their conclusion on theoretical analysis of Pastry and some experiments on it. They almost considered all issues and experiments' results are convincable. I just want to ask all the charateristics are because of just Pastry design or because of some basic charateristics of overlay networkdesign , such as : make use of internet proximity knowledge? Most interesting ideas: I am very interested in Pastry's routing table,message routing rule, which are of most significance to get to know why Pastry had produced such experiments' results. Weakness&Questions: Maybe authors can show us more detailed relation or general charateristics among all overlay networks? Do other overlay networks making use of network proximity show similar performance as Pastry? Comments: Anyway, this is a good paper. It depicted a very clear and detailed view about the network locality property of a peer-to-peer overlay network. ############################################ ############################################# Date: Fri, 3 May 2002 07:01:58 -0500 (CDT) From: Ivona Bezakova Resilient Overlay Networks [ABKM01] 1. State the main contribution of the paper: The paper proposes resilient overlay network (RON) architecture for fast detection and recovery from various network problems (e.g. failure, congestion). The idea is to have a small set of nodes which constantly ping themselves and thus they monitor the underlying Internet paths - if a problem arises, a packet is routed through a different RON node. 2. Critique the main contribution. a. Rate the significance: 4. The possibility of selecting client's own metrics, membership mechanism, etc makes the protocol attractive. b. Rate how convincing: 4 - well-written 3. What are the three strongest and/or most interesting ideas in the paper? - Selection of client's own metrics. - The RON protocol is a RON client itself. - Classification / Policy Routing. 4. What are the three most striking weaknesses in the paper? - Failures of the links from clients to the RON nodes. - What happens if there are many different metrics? Doesn't it degrade performance too much? (Do I understand correctly that a new metrics may ask for a completely new attribute to be monitored by the RON nodes?) - Malfunctioning nodes are dealt with on the human level. 5. Name three questions that you would like to ask the authors? - I did not understand the reasoning (if any) behind the score formula 6. Detail an interesting extension to the work not mentioned in the future work section. 7. Optional comments on the paper that you'd like to see discussed in class. - See 4. Exploiting network proximity in peer-to-peer overlay networks [CDHR] 1. State the main contribution of the paper: The authors evaluate effectiveness of proximity-based Pastry system. Their evaluation includes (1) a theoretical model based on uniform distribution of key-requests and source nodes - they compute the expected number of routing hops; (2) a simple model of a p2p system: uniformly distributed nodes on a sphere; (3) experimental results using three different underlying topologies - sphere, GATech (transit-stub), and Mercator (based on real measurements of the Internet and hierarchical routing). 2. Critique the main contribution. a. Rate the significance: 3-4. I think that study of this type is important because when the proximity comes into picture, most theoretical guarantees are not valid anymore. On the other hand I miss conclusions following presentations of the statistics. b. Rate how convincing: 4-5. I really like their use of sphere model to validate correctness of the simulator. I think that this should become standard whenever applicable. 3. What are the three strongest and/or most interesting ideas in the paper? - Theoretically computing expected values of the studied parameters on a simple sphere model and using this to verify correctness of their simulator. 4. What are the three most striking weaknesses in the paper? - Missing simulator conclusions.