=================================================== From takhoa@owlnet.rice.edu Wed Jan 21 17:25:10 2004 =================================================== The paper makes the argument that packet-switched networks are inferior to circuit-switched networks. I find the argument for circuit-switched problematic. First of all, telephone networks are setup by a few major entities. Circuit-switched networks require extensive planning and coordinations. This was not feasible with the Internet architecture where small segments of the network are built and connected together at different times. Circuit-switched networks are only highly tolerant because the few operating entities are willing to spend resources on keeping them reliable. Packet-switched networks can exist in a more distributed and heterogeneous fashion because of its distributed nature. I also find that aside from offering some interesting statistics about the utilization of different networks, this paper hardly offers any new idea or suggestions. The authors' vision of a clean-slate start seems to be very close to what already exists. Currently, circuit-switched networks are already in place between major routers running IP over SONET. With the bursty and assymetric nature of the internet traffics, I think packet-switched are still better in most instances, unless for major communication channels where traffics can be aggregated for a common end point. =================================================== From dushu@cs.rice.edu Wed Jan 21 17:40:33 2004 =================================================== This is an interesting paper to read. In this paper, the authors questioned the popular claim that IP is going to dominate the world of communication, furthermore, they even suggested that circuit switching network should replace IP switching in the backbones of the Internet. The authors built their belief based on the observations of the several problems IP now is suffering, for example, the low utilization of the IP network, the difficulties of the routing management and the lack of the QoS support etc. Although their observations of the problems which IP is having now is kind of right, I think the conclusions they draw are wrong or biased. Basically, the problems they mentioned are not all because of the IP switching mechanism, at least,not inherent from IP switching. Most of them are coming from the historical reasons, like the lack of the QoS support or complexity of the routing. As to some other blame of IP in this paper, like the inefficiency of IP, is much more misleading, just as what the authors did in most of the comparison between the IP switching network and the circuit switching network. As to the details of this, Johnny and I will present them tomorrow. review of [Cla88] It is a very good paper to read. The author introduced the historical design motivations behind the current TCP/IP network and he also discussed the the priorities of these design goals, therefore explained to us why the TCP/IP is like what we are looking at now. After that, the author introduced some inherited problems the current IP is having and suggested a possible way to solve them in the future. =================================================== From gulati@cs.rice.edu Wed Jan 21 21:18:27 2004 =================================================== Is IP going to take over the world (of communications)? This paper discusses some of the common assumptions about IP and argues that some assumptions are not supported by facts. Although packet switching has some desired features like ability to do statistical multiplexing, routing around failures, easy maintainability, simple implementation, its not clear that packet swicthing will replace circuit switching in future. The authors raise doubt about the usefulness of these features and compares them to characteristics of the circuit switching. Some of the argued assumptions and reasons against their credibility are: 1. IP dominates global communications: Market shares of television and telephone industries are much higher than share of internet. 2.IP is more efficient: IP networks are over provisioned and utilization is kept low to reduce latency. 3. IP is robust: A single router can cause lot of damage to the network service and IP networks take a long time to recover. 4.IP is simpler: IP interacts and uses lots of other protocols to function. This interrelationship makes it complex to handle and simplicity does not scale. 5.IP can support real time applications and telephony: Its not clear if current approches for QoS and real time guarantees can economically compete with existing telephony infrastructure. Finally the authors conclude that IP needs to coexist with other circuit switched technologies and work with them. They also state that even if we start with a clean slate, we will use circuit switching at the network core and packet switching at the edges of the network. =================================================== From twngan@cs.rice.edu Thu Jan 22 07:59:05 2004 =================================================== This paper argues that while IP (and packet switching in general) is good for exchanging data and best-effort data services at the edge of the networks, the core of the network should use optical circuit switching as a platform. The authors claim that most, if not all, the advantages people thinking of IP and packet switching are just misbeliefs. Yet instead of claiming that circuit switching should displace packet switching, the authors just argue that only the core of the network should be displayed by circuit switching. While I agree that certain part of the packet switching network could benefit from ideas of circuit switching networking, I disagree with the authors' attack on IP. In particular, the authors blame IP because of the complexity of the addressing and routing issues, which is yet to solve even for circuit switching networks if they were replacing packet switching networks. Also similar is the complexity of the routers due to VPNs, security, access control, etc., which is also needed for circuit switching networks. The authors avoided the significance of historical and social factors (like mandated by government of performance guarantee and better planning for telephone networks). Last but not least, the downplay of statistical multiplexing is ridiculous, since not all services requires the best possible response time, and the performance/cost tradeoff should be on the hand of the users based on their needs. And even telephone networks assumes statistical multiplexing and try to benefit from it. =================================================== From animesh@cs.rice.edu Thu Jan 22 14:59:43 2004 =================================================== Review ( Is IP going to take over the world of communications) -------- This paper advocates that IP will not completely take over the world of communications. It argues that packet-switched IP should only be at the edge network with circuit-switch comprising the core network. It bases this cnclusion by first exposing some of the myths of IP. IP market is in comptetion with other worldwide markets like TV, Cable, Phone which will continue to exist and in no way currently dominates the communication market. The second myth about using IP because packet-switched network utilizes scarce bandwidth effciently although true, is no longer of importance because the bandwidth at the core of the network is no longer scarce. Bandwidth is greatly overprovisioned for several reasons, infact one of them also being because the packet-switched network performs unstably and badly when the network utilization is high. Another myth surrounding IP that it is more robust has been conteneted in this paper. Although true, that IP can route around failed componenets it has been studied that the recovery time for circuit-switched network is lower as compared to the recovery time in IP resulting from high convergence times in the routing protocols. Another myth is that IP is simpler, although the reality is that the complexity in the IP router software and hardware is much more complicated that complexity of circuit-switched hardware/software. The other myth surrounding IP is that one day it can support all real time applications include voice over IP. Although efforts are being made to support QOS guarantees in IP, the guarantees cannot compete with the guarantees that can be provided by TDM circuit-switched network. IP if it takes over completey would require it to be compleety robust, stable, managaeable and have ability to give QoS guarantees. As of now, IP does not excel in these areas leading us to conclude that circuit-switched network would continue to be in existence with IP which would primarily operate at the edge of the network. =================================================== From anwis@cs.rice.edu Thu Jan 22 15:14:16 2004 =================================================== Is IP going to take over the world (of communications) ? This paper discusses the common misconceptions people have about IP and analyzes why IP is not going to be the future of networking (contrary to popular belief). The second part of the paper especially discusses how IP fits into the future of networking. The common assumptions are the following: 1) IP already dominates global communications By pure statistics, the authors show that this is plainly not true. Hard to argue against. 2) IP is more efficient The key points here are that network administrators over provision their networks (so as to render statistical multiplexing, a key feature of packet switched networks, useless) and that variance of network utilization is a very important metric for users to be happy. Taking both of the above facts into account, along with the fact that variance is very difficult to measure in a packet switched network implies that circuit switching seems to be more relevant given present technology. They also ran a correspondning experiment to show the user latency for packet switched and circuit switched networks. 3) IP is robust The authors cite a study that show that the mean downtime in phone networks is 5 min/year compared to a 471 min/year for computer networks. Moreover, the fact that control messages are subject to the same failure rates as data messages in computer networks means that inconsistent routing states may be reached and the resulting user traffic may be disrupted. Ironically, it seems that packet switched networks are less reliable than circuit switched networks because of these problems. 4) IP is simpler This section argues that packet switches are not simpler than traditional circuit switches based on the number of lines of code for a typical router operating system and the hardware in the forwarding path of each. The argument is made concrete by giving the number of gates for the linecards, a good measure of complexity. Moreover, the authors argue that optical switching technology is the future and circuit switches can adapt to this much better than packet switches. 5) Relationship between IP and real time applications and telephony The authors argue that end to end of quality of service is extremely challenging and the problem has not been solved yet. Moreover, they argue that it is difficult to imagine users switching to IP based networks for telephony given that current infrastructure provides the service better than IP could ever hope to at almost no cost to the user. 6) Future of IP The authors argue that dependability is the driving force behind any technology that is to become the universal infrastructure. But, IP cannot provide sufficient robustness, stability, traffic isolation, traffic engineering, fault isolation, manageability, and quality of service. Also packet switching will be more dominant at the edge of the networks whereas circuit switches will take advantage of optical technology and become dominant in the core of the network. As seen before, circuit switching may actually be more reliable and robust than packet switched networks. Moreover, packet switched networks at the edge will share the circuit switched backbone at the core with other technologies such as voice traffic. The main argument is that packet switched networks may take advantage of available bandwidth at the edge because its scarce and then multiplex traffic onto the circuit switched core. At the core, circuit switches have proven to be robust and may take advantage of optical switching technology to drive down costs. =================================================== From mittal@cs.rice.edu Thu Jan 22 15:16:20 2004 =================================================== Paper - Is IP going to take over the world(of communications)? The paper believes that there is more confidendce and hype over in IP in the scientific community than it deserves. The authors start off by suggesting what they believe are the common misconceptions about IP, then go about refuting these one by one. According to the authors, IP does not dominate global communications as of today - Internet is currently only a small fraction of communication structure. Also, it is not as efficient as portrayed -- network utilization is low nowadays and it is no longer the case that links are slow, congested and expensive. IP is not robust to failures and might lead to inconsistent routing states. With circuit switching, we transmit messages over a separate channel and routing is much simpler . Also, it would be difficult to support telephony and other applications that require real-time support over IP, even after years of research, we are still not ready to build such real-time applications over IP. Looking at the ecnomic perspective, there are few financial incentives to carry voice over the Internet, for example. The paper concludes by saying that if we proceed with a clean state now and were given the option of choosing which technology to use, it would make sense to use packet switching in scenarios where bandwidth is scarce and in cases where the sources are heterogeneous and disparate. However, at the core of the network, it would be wise to use circuit switching due to its robustness, simplicity and being able to provide higher guarantees. The paper might have been a little harsh on IP, but I generally agree that in future, we might have to employ techniques which provide more guarantees to the end-users. IP would continue to be used, however. According to me, the main point that the paper makes is IP is not as efficient as circuit switching, as links are underutilized making IP less effective, and also IP's inability to provide real-time guarantees would prevent it from being universally acceptable. =================================================== From santa@cs.rice.edu Thu Jan 22 15:30:05 2004 =================================================== Is IP going to take over the world (of communications)? The growth and success has given rise to the assumption that IP is here to rule the world of communications. The authors debunk this folklore by striking down many of the assumptions that is expected from IP datagram network. They argue that if the internet was started with a clean state, the same model - packet switched routers interconnected by circuit switched transport network - would still be the best. In striking down the IP folklore, some of the sacred cows of networking are revisited. "IP already dominates global communications" - this assumption is proved wrong by figures which show that the circuit switched network is a much bigger market. While IP is theoritically more efficient in the use of bandwidth than circuit switching (which was important earlier because of low-bandwidth links), current low level of utilization of the internet point to the fact that efficiency is not important any more. While efficiency was once a critical factor, it is outweighed by the need for predictability, stability and low delay, which force networks to be run at low-utilization thereby forfeiting the benefits of multiplexing. Robustness of IP was a key goal of the DARPA protocols, but the uptime of internet does not even come close to the phone network uptime. The IP network can get to an inconsistent state much faster, debunking the notion that packet switching is inherently more robust. Complexity should be kept at the endpoints was one of the guiding goals of DARPA protocols. But the implementing a high performance router at very high speeds with many added complexities introduced (like multicast, QoS, security) is extremely challenging giving rise to the notion that IP's simplicity does not scale. Support for real-time applications over IP can only be done by either heavy overprovisioning which is self-defeating, or by IntServ or DiffServ QoS which have still not seen daylight despite extensive research. The authors propose that lots of research needs to be done in IP to make it more robust. Also, packet switched IP networks would have to coexist with high-speed reliable circuit-switched core. =================================================== From amislove@rice.edu Thu Jan 22 15:41:26 2004 =================================================== In this paper, the authors present an argument to dispell the myth that Internet Protocol (IP) is going to be the dominant communications infrastructure. Specifically, the authors argue that packet-switched networks are appropriate for certain communications platforms (such as web traffic or file transfers), but not for others (such as real-time applications). The authors base their position on the facts that IP is not as robust as circuit-switched networks, IP is more efficient (but only because networks are overprovisioned), IP is much more complex and hard to manage, and that IP is inherently inappropriate for certain communication patterns. Finally, the authors discuss what could be done if we could redesign the Internet from scratch, and decide that packet-switching is only appropriate for the end routers, while the core and applications which have QoS demands should be served by circuit-switched routers.