dedicated to discussion of ideas for a new Internet architecture
It’s nice to see the press taking an interest in GENI and FIND. The Associated Press released an article on future internet design yesterday, one that was featured on the website Digg:
http://www.signonsandiego.com/news/business/20070413-1017-rebuildingtheinternet.html
Hi all, I thought I’d post my final project for the CSE570 course. I’m doing this because I think it’s not too shabby 
, but also because I’d be interested if other members of Class CSE570 of ‘06 were to post theirs. I’m sure it makes good reading, but on a more selfish note I had to leave before the 18th so I missed presentations on that day. So to kick off the dare, here is mine: http://www.cse.wustl.edu/~harri/stuff/SimulationMetanet.pdf
Read the rest of this entry »
Any communication process can be divided into one of the three main categories, unicast, multicast, and broadcast communication. In unicast, messages are sent from one source to one destination; the message may traverse through several intermediate nodes. In broadcast, messages are sent from one source to all destinations which have physical connectivity to the source. Multicast is the process of sending the message from a single source to a subset (containing two or more hosts) of all physically connected destinations. When messages have to be sent from one source to multiple destinations, multicast is generally a much more efficient mode of communication than simply unicasting the message to all destinations. Multicast is efficient because the message need not be replicated for every destination; replication only occurs when the message gets close to the destination. Thus, the links which are close to the source do not require large bandwidth, which is otherwise needed if the same message is unicasted multiple times. Clearly, multicasting has several applications, some of which are multi party video conferencing, broadcasting information to a large number of users, real-time video distribution, etc.
Read the rest of this entry »
IP multicast is a must for the Internet. It provides many advantages over a plain unicast network. With demand growing toward more streaming media, IPTV, and video conferencing applications, multicast becomes more and more of a necessity. There are many advantages that multicast offers to these type of applications. The infrastructure for IP multicast already exists. It is supported by Ethernet and many multicast protocols are already implemented in routers.
Read the rest of this entry »
A multicast communication service sends packets from a source to a set of destinations, also called multicast group. The basic underlying idea is to propagate the packets into the network so to reduce the bandwidth involved. If, for instance, a packet has to be sent from a source on ISP x to N recipients connected to another ISP y, then a unique copy of the packet will be first sent from ISP x to ISP y, and then ISP y will locally dispatch N copies of the packet to the interested recipients. In a more general scenario a multicast dissemination tree, that is, a minimum spanning tree rooted at the sender, will be used in order to determine how a multicast packet will be propagated in the network. This solution will involve a degree of bandwidth consumption which is far less than simply having many unicast transmissions between the sender and every recipient.
Read the rest of this entry »
Streaming media applications such as real-time video and telephony continue to grow and become a large component of Internet traffic. Since such applications favor low delay over reliable transmission they often opt for the unreliable transport protocol UDP. However, UDP lacks built-in congestion control mechanisms leaving applications on their own to implement congestion control, if at all. The other alternative to UDP is TCP which provides congestion control along with reliable transmission. However, applications tend to avoid TCP because delivery guarantees can delay packets to the point that they are unusable to the application and better off dropped. Without proper congestion control, high-bandwidth streaming applications will create havoc on the Internet.
Read the rest of this entry »
As a network application developer, you have only two realistic choices for your application’s transport layer. You could pick UDP (User Datagram Protocol), which provides unreliable, connectionless data transport, for applications where timeliness is the primary data-delivery concern. You’d be forced to implement congestion control yourself, and one bug could render the network unusable.
Read the rest of this entry »
This paper summarizes the design of Datagram Congestion Control Protocol (DCCP). DCCP is a congestion control protocol to be used with unreliable transfer protocols such as UDP. There are many applications such as streaming media or video conferencing that prefer timely data instead of reliable data. If these applications had the choice of either retransmitting a lost packet or transmitting a new packet, they would choose to transmit the new packet. By the time the old packet would arrive, it would be outdated and useless. In these cases, TCP is a poor choice because of its extra effort given to reliability.
Read the rest of this entry »
The question I want to address is the following: which, if any, would be the advantages of having geographic addressing in WANs? In order to analyze this problem, I will first summarize how routing is performed on WANs, what geographic routing is and in which context it has been deeply studied. The objective is to find out whether some of the requirements which motivated the idea of geographic routing apply also to WANs.
Read the rest of this entry »
Network addressing and routing protocols have received enormous attention since the inception of the Internet. Any addressing scheme used in the Internet must serve three fundamental objectives: identity (so that end nodes can be identifiable), location (so that packets destined for the end nodes can be routed), and reachability (which links should be taken to route messages from one node to the other, note that in the current Internet, by default, every connected node is reachable).
Read the rest of this entry »
The routers on today’s Internet forward packets based on the layout of the IP address, in a method called Longest Prefix Matching (LPM). Unfortunately, LPM becomes harder as line speeds get faster. Direct lookups are one method for LPM and complete in O(1) time, but would require an exorbitant 17GB of memory. Trie representation enables a tradeoff between memory and lookup speed, but is held back by the glacial pace of memory latency improvements, making trie lookup a challenge for rates greater than 10 Gbps. Ternary Content Addressable Memories (TCAMs) enable O(1) associative lookups, and thus line rate speeds, but are expensive and power-hungry. What if there was a method of addressing that could enable line-rate lookups with minimal memory, power, and cost?
Read the rest of this entry »
Mobility is becoming a huge driving force on the Internet today. More and more mobile users connect to the Internet everyday through the use of laptops, PDA’s and smart phones. Applications that take advantage of this shift towards mobility are gaining popularity. It is also becoming apparent that having host location information available in the network can unlock a whole other class of applications. Already, applications exist that could benefit from the addition of location information to the network. Geographic location in WAN addressing would reach this goal of location information in the network to enable new services. In addition, geographic addressing has the added benefit of simplifying routing.
Read the rest of this entry »
Despite the fact that IP Multicast has been widely studied in the last fifteen years, none of the proposed infrastructures to implement this service model has been widely deployed and has been proven convincing. In fact, it is still under question whether the benefits implied by the introduction of IP Multicast would compensate and overreach the complexity of its deployment and management in the network level. However, the authors notice how several applications could largely benefit from an underlying multicast communication service, and how the spreading of such applications is dramatically increasing. It is the case, for instance, of multiplayer games, Internet TV technology, video conferencing, file sharing, software updates and more. Thus, the paper revisits IP multicast and proposed a novel approach, called Free Riding Multicast (FRM) for its deployment.
Read the rest of this entry »
The authors revisit one of computer networking most studied ideas, IP Multicast. Multicast is one of the few ideas that many technical papers proposed ideas for implementation, but hardly any real world, wide-use protocols have been deployed. The complexity and scalability of the multicast idea is what has caused the difficulties. The authors argue that there are enough applications that would benefit from it and maybe we should take another look at multicast.
Read the rest of this entry »
This paper explores network anycast as an overlay system. When I typically think of anycast, I imagine IP level anycast. Traffic sent to a specific IP address is routed to the “best” of many possible hosts associated with that address. Best could mean closest or least loaded depending on the network. However, the OASIS system presented in this paper does not enable this sort of IP layer multicast. Instead, OASIS provides anycast at higher layers, in particular through the use of DNS and HTTP redirects. In this way, I was disappointed once start into the paper. OASIS, as it turns out, is more of a server selection service than IP anycast.
Read the rest of this entry »
As people finally realize the threat of denial of service attacks, a trend towards distributed systems is appearing. For websites, services such as Akamai have become popular, even though they have a higher monetary cost than a company hosting their websites on their own. Services can be distributed across vast geographic areas, not only to assure that there is no one point of failure, but also to decrease latency, such that users can find replicas of services near them. OASIS implements this as a distributed anycast system, based upon geographic mapping, which also provides answers to problems inhibiting past anycast efforts.
Read the rest of this entry »
Enabling Contribution Awareness in an Overlay Broadcasting System, by Yu-Wei Sung, Michael Bishop, and Sanjay Rao.
Presentation by Brandon Heller.
You are the owner of a small business and you are looking to advertise via email. You, like everyone else, hate spam mail so you do not want your advertising to be spam. There is such a great opportunity to advertise via email that you comply with CAN SPAM, which is a bill passed by Congress that list requirements to send unsolicited advertisments over email. This bill was signed into law to cut down on the amount of spam. So, you comply with CAN SPAM and your employees do a great job developing the advertisement. Everything is set to go. You send your email, but to your surprise it is not delivered to anyone on a major Internet Service Provider (ISP). You wonder what happened and decide to look into it in order to avoid this in the future.
Read the rest of this entry »
In Romeo and Juliet, Juliet wonders ‘What’s in a name?’ Earlier, she also asks ‘Wherefore art thou Romeo?’ While these may be appropriate queries when speaking of the names of persons, in the realm of the Internet, there are many issues with naming, and it is not much ado about nothing. In the real world, names are of less importance, as we can easily distinguish people by their characteristics, even those persons with the same name. Such differentiation is much more difficult with the Internet, leading to many restrictions, as precise identification is a necessity. One’s known alias, a unique hostname, is resolved to the true identity, a unique IP address, by Domain Name Servers (DNS). And therein lies the problem, for this naming architecture can be abused in numerous ways, from thieves and con artists trying to masquerade as official businesses, to attackers attempting to amplify Denial of Service (DoS) assaults. The remainder of this essay will detail the numerous flaws with the current Internet naming system, demonstrating that once again, a common IP mechanism remains only out of necessity, rather than its greatness.
Read the rest of this entry »
There can be no doubt that human-understandable names are necessary for the Internet to be useful to the broad population. After all, the vast majority of people need something easy to remember, like ‘www.yahoo.com‘, instead of some string of random-looking numbers. Having names also allows us to change the underlying mapping, as when a machine moves around, without affecting how the end users gain access to that machine. So, given that we want to attach (hopefully) meaningful names to entities on the Internet for ease of communication, we have to decide how to go about doing so. In this essay, I will argue that a single globally consistent system should be used.
Read the rest of this entry »
The Domain Name System (DNS) is the globally distributed database that maps human-readable domain names like “www.microsoft.com” or “wikipedia.org” into router-readable IP addresses such as 207.46.199.30 or 66.230.200.100. In fact it keeps a whole set of typed resource records for each domain name, such as the address of the mail exchange server for a domain or the reverse mapping from an IP address to its canonical (fully qualified) domain name. A large number of important Internet protocols, such as HTTP (the Web) and SMTP (e-mail), rely heavily on the global namespace served by DNS. Without DNS to map the “arl.wustl.edu” part of “http://arl.wustl.edu/~jst/reInventTheNet/” onto a routable address or the “cse.wustl.edu” part of “harri@cse.wustl.edu” to the appropriate e-mail server, the usability of today’s Internet would be much reduced. A later and interesting benefactor is the XML namespace extension [1], which uses DNS domain names primarily as a mechanism for disambiguating XML document typenames, while also exploiting the fact that the names point to hosts that can serve up the corresponding type definition files. Similarly, the Java package mechanism uses the same trick to preclude name clashes in the Java type hierarchy namespace.
Read the rest of this entry »
Internet QoS is one of the most comprehensively studied topics among the research community. One of the primary reasons that QoS has got such an attention is that the current Internet is unable to provide a wide range of QoS policies which are needed by several emerging real-time applications. For instance applications like VoIP, real time streaming media, etc require both bandwidth and delay guarantees. The performance of these applications can deteriorate considerably in face of bursty losses, and time varying network delays therefore it is important that the traffic from these applications are isolated from other traffic and from the time varying congestion in the network.
Read the rest of this entry »
OverQoS is an overlay-based architecture for improving the best effort service of the current Internet with minimal cost. The paper describes the design, implementation, and evaluation of their architecture with very promising results. They use what they call a Controlled Loss Virtual Link (CLVL) to bind the loss rate. With this OverQoS architecture, it is able to smooth bursty packet losses, prioritize packets, and achieve loss and bandwidth guarantees. The authors demonstrate their architecture on two real-world applications, RealServer and Counterstrike.
Read the rest of this entry »
This paper describes an overlay-based architecture for providing quality of service (QoS) guarantees in the Internet. It demonstrates a prototype implementation and evaluates the prototype with real applications.
The authors correctly note that the reason why QoS schemes have failed to be adopted in the past is that they require all transit nodes along a network path to implement QoS mechanisms. This runs directly into our well-known “Internet Ossification” problem: you can’t coordinate an upgrade among all the diverse players. Therefore, the authors explore techniques for providing QoS without universal deployment by using an overlay. Effectively, they deploy a QoS scheme over some fraction of the nodes in the Internet.
Read the rest of this entry »
Overlay systems have become a popular means of broadcasting on the Internet. Without native support for multicast, overlays have filled the functionality gap in IP. However, as overlay broadcast systems become more popular, it is ever more clear that such systems are more and more characterized by heterogeneous clients with varying, asymmetric bandwidth. Traditional overlay broadcast systems break down under these conditions because they lack the mechanisms to cope with this varying levels of participation resulting in inefficient resource utilization and inequitable performance. This paper takes a look into extending popular overlay broadcast systems onto heterogeneous, mixed environments in a way that is both efficient and fair to users.
Read the rest of this entry »
The paper proposes an overlay network using Forward Error Correction (FEC) along with retransmissions to reduce and / or shape packet losses, trading off bandwidth for lower loss rates. The intent is to improve QoS without requiring changes to existing routers, relying on overlay router nodes to implement the new functionality.
Read the rest of this entry »
In this paper, the authors aim to develop a broadcasting system that can distribute content from a source to hosts while considering the heterogeneous capabilities of the hosts in the system. There are two main motivations for the paper. The first one is the requirement to handle heterogeneous properties of hosts, which is a result of the differences in the outgoing link capabilities of different hosts. The second one is the need to handle cases where not all hosts can get the content at full rate, because of the limitations on the bandwidth. For these purposes, the authors propose a system where hosts receive content at a rate in proportional to their contribution to the distribution of the content.
Read the rest of this entry »
The Internet was introduced as a government funded research project. The ancestors of the Internet, ARPAnet and NSFnet, were used to connect research institutes and government computing centers, and they were not allowed to be used for commercial purposes [1]. In the late 1980s, commercial network providers which are called Internet Service Providers (ISPs) have emerged. Most of the investment in routers, servers and infrastructure came from ISPs with the commercial usage of the Internet [2]. Since then, there has been a great progress in the quality and the range of both the infrastructure types and the services. Backbone speed of NSFnet was initially 56 Kpbs, whereas that speed is offered at the slowest type of connection, dial-up connection, today. Improvements in computational hardware and transmission media have an important contribution to this fast development, but the most important factor enabling the development is definitely the demand from the network users. The demand for better Internet access is so high that policies regarding broadband access have been a debate topic in presidential campaigns [3]. As a result of this demand, there is a big potential revenue in this still immature market, and the result is the competition among several network providers for sharing this revenue.
Read the rest of this entry »
Imagine sitting in front of your computer getting ready for another episode of surfing the web. You open your favorite web browser and wait for your home page to load. To your surprise your homepage does not load; instead its another webpage. You check the address bar, but it indicates you are at http://www.myhomepage.com. You think you must have picked up some spyware and it is affecting your homepage and disguising your address bar so it appears you are on the correct page.
Read the rest of this entry »
The Internet access is still not cheap even though the users are constantly presented with unwanted advertisement and emails. There are still severe complains about how hard for customers to cancel their service from some particular network providers. All these phenomena (broader level of uses services) can be improved by introducing real competition for network providers or internet service providers.
Read the rest of this entry »
Active Networks are definitely a computer scientist’s take on what networking could be. Scott Nettles of UT Austin and I wrote up some history and context in a paper, reachable at: http://repository.upenn.edu/cis_papers/46/ and give some visions for futures. While the name “Active Networking” evokes a variety of responses in the networking community, I believe the reality is far more compelling than the perception.
Best,
-JMS
GENI has revived interest in virtual infrastructures. Some of the proposals for active networking - see http://www.cis.upenn.edu/~jms/switchware.pdf - included applications which would be realizable with a programmable virtual infrastructure. The applications in Section 3 of this technical paper from 1996 were far-fetched at the time but illustrate both the variety and capability achievable from a distributed systems view of networking.
Key questions: (1) are these (or similar) applications still hard to realize? (I think so, but am always eager to be educated), and (2) how would they (or similar) applications be realized with GENI.
-JMS
Current Internet architecture has several limitations and to a large extent, these limitations stem from the structured addressing scheme used in the Internet. The Internet uses IP addresses to represent both the location as well as the identity of the nodes, and this design choice has been motivated partly by two reasons: i) when the Internet was developed, most of the participating nodes were static in location, therefore their location and identity were more or less the same, and ii) The hierarchical structured form of the IP addresses and the location information embedded within them, enables much scalable routing, and millions of addresses can be handled efficiently (the routing tables at each router remains small and easy to maintain).
Read the rest of this entry »
This paper attempts to address a specific problem in anonymous routing, that of node failure, which can lead to data loss, jitter, and even degradation attacks that reduce anonymity. This approach uses the standard technique for anonymous routing, where a path is randomly assigned by the sender through a series of overlay nodes, but substitutes a group of nodes for each single node in older approaches. This paper does not seek to improve substantially on the anonymity of prior work, but merely to provide increased resilience to network dynamics and node failures.
Read the rest of this entry »
It has been generally recognized that using IP addresses for both identity and location is not a good thing. Many proposals for separating the two exist, most often in the context of disruptive changes to the Internet. In this paper, the authors give a different view point on the problem. Instead of merely separating identity and location, they discard the notion of location entirely. More accurately, they propose a system where there is no explicit representation of location and routing is done on identity. Of course, in many ways this is similar to what is done with IP, as addresses are used for identity. The real departure from the current system is that the routing label is flat, i.e., there is no structure to the labels (identities). This challenges the long held belief that the structure of routing labels (in this case, IP addresses) allows the act of routing to be scalable, and that is really what this paper is all about.
Read the rest of this entry »
Rolling on the floor laughing. That’s what we normally think of when we consider the acronym ROFL. However, when we consider the current usage of IP address for both identity and location, it is no laughing matter. This current usage causes many issues, including difficulties with mobility and tenuous definitions of user identity. This paper provides a new scheme to fix this problem, entitled Routing on Flat Labels (ROFL). Location is removed entirely; instead, routing is based upon the unique, constant identifier of a user. This builds on top of DNS, requiring no new infrastructure, while making identity allocation and access controls easier to implement. The paper proceeds to demonstrate how not only is this new method beneficial, but also scalable to the Internet as a whole.
Read the rest of this entry »
This paper presents a new architecture called Cashmere, which enables anonymous routing, and provides both source anonymity and unlinkabality of source and destination. The key benefit of Cashmere overtraditional approaches is that it provides an increased resilience to node failures and node churns which generally degrades the performance of traditional anonymous routing protocols based on Chaum-Mixes. Traditionally Chaum-Mixes based routing protocols achieve anonymity by relaying the traffic through a sequence of nodes, such that any two nodes, which are not adjacent to each other along the path, are unable to identify each other. Thus, if the relayed path contains more than two nodes, then there is no way the destination can identify the source. More specifically, no downstream node can identify the upstream nodes.
Read the rest of this entry »
In the current Internet architecture, IP address is used to define both the identity and the location of a host. This causes problems in many applications including mobility, multihoming etc. The authors propose a new architecture to separate identity and location. This proposal suggests using flat labels for routing, which only defines identity. Location of a host, which defines the connection point of the host in the network, is discarded altogether. Labels are selected to be flat, meaning that they contain no semantic information. After explaining the reasons for selecting this way, the authors set their goal as designing a routing mechanism to show the feasibility of this approach.
Read the rest of this entry »
The paper describes how the structure of prefix-routing DHT’s may be used to enable a source to securely send anonymous messages through a chain of intermediaries, such that each intermediary is in fact a group of nodes in the DHT. This redundancy ensures that the failure of intermediary nodes is unlikely to break an anonymity forwarding path, substantially increasing the expected lifetime of such paths, which improves reliability and performance.
Read the rest of this entry »
Designing DCCP: Congestion Control Without Reliability, by Eddie Kohler, Mark Handley and Sally Floyd.
OASIS: Anycast for Any Service, by Michael J. Freedman, Karthik Lakshminarayanan and David Mazières.
Revisiting IP Multicast, by Sylvia Ratnasamy, Andrey Ermolinskiy and Scott Shenker.
There is a fairly recent trend toward so-called “smart networks,” where advanced features – QoS, security, application session management – is moved into the network infrastructure. This is obviously counter to the basic principles o the Internet: keep the network as simple as possible and implement other functionality at the end point. However, let’s not reject the concept on this point. There are so many other, better grounds!
Read the rest of this entry »
One of the fundamental design principles of the Internet is the end-to-end arguments [1]. The Internet can be seen as a structure that is formed of the core infrastructure of interconnected routers responsible for data transfer and the end systems attached to this core infrastructure. The end-to-end arguments state that the core infrastructure should not contain any specific application-level functions, it should carry data packets the same way irrespective of the type of application the data belongs to. Applications can run on end-systems using this generic structure and implementing optimizations at the application layer. One of the main reasons for the assertion of these arguments is that it is generally not possible to support the requirements of the applications at the core alone. Even if some functions can be added to support some specific applications, there is always the danger of these functions hindering the operation of other applications. These functions added to core would also increase the complexity of the core, which in turn can make the core less robust and more susceptible to failures. The end-to-end arguments state that, as a result of these potential dangers, if a function can be implemented at the application layer, it should not be a part of the core. The evolution of the Internet showed that this basic design principle is one of the most important factors in the innovation of new applications.
The Internet architecture was primarily intended to provide unicast communication between pairs of fixed endpoints. However, a high spectrum of applications could benefit from different communication services, like multicast and anycast, and from host mobility. The goal of this paper is to propose an overlay network introducing a level of indirection which allows an easy implementation of the above paradigms. This overlay, called i3 (Internet Indirection Infrastructure) provides a rendezvous-based communication abstraction which decouples senders from receivers.
Read the rest of this entry »
Although proposals for IP mobility, multicast, and anycast have led to Internet standards, and even router implementations, these beneficial ideas have yet to see widespread deployment. These proposals prioritize application stability over network stability - reasoning that we should keep good old TCP,UDP, and everything on top of them unchanged, even if it means that we have to change every router on the internet. The distributed nature of the Internet means all ISPs, or at least a subset of the largest providers, must choose to deploy a proposal before anyone sees a benefit. Thus, deployment is the main obstacle of such proposals.
Read the rest of this entry »
The Domain Name System is yet another piece of the Internet that needs to be re-examined in the light of the current world. The design of DNS has allowed it to cope reasonably well with the massive growth of the Internet, but there are clearly problems that can’t easily be overcome just by tweaking the system. Recently, there have been a number of attempts to overcome these obstacles by designing replacements to the DNS, and this paper represents one of them. The Cooperative Domain Name System (CoDoNS) overcomes some of the problems with the legacy DNS while improving performance (latency) over DNS. It has been implemented on PlanetLab and is available for use as of this writing.
Read the rest of this entry »
Today’s Internet was built on the architecture to provide unicast communication between stationary locations. This works great for e-mail, but what about massively-multiplayer online role-playing game like the popular World of Warcraft or broadcasting a live video to cell phones moving all over the world. Another big trend is the use of Voice Over IP on cell phones. This would require the cell phone to work as a server as well as a client. The current Internet architecture does not do a very good job at providing these anycast, multicast, broadcast and mobile services. The paper “Internet Indirection Infrastructure” proposes generalizing the Internet’s point-to-point communication in order to better provide multicast, anycast and mobility services.
Read the rest of this entry »
The Design and Implementation of a Next Generation Name Service for the Internet
Domain Name System (DNS) is an essential part of the Internet. The authors had come up with a new DNS architecture to reduce the problems of the legacy DNS. Even though the analysis had provided overall good reasons to demonstrate that the new architecture is better, I found some good and bad arguments from this paper.
The legacy DNS is slow by the authors’ claims, however the core functionality of DNS is for lookups and translations. In my opinion, as along as most of the popular and time sensitive websites are readily accessible through either Akamai technology or the decreasing routing time of the Internet, the slowness of the legacy DNS is well tolerable. Because the legacy DNS is like a dictionary, currently more static and we will normally use just a certain percentage of its information. With the same reason I will argue about the problem of implementation errors for the legacy DNS.
Read the rest of this entry »
This paper examines the current Internet Name Service (DNS) and finds it lacking. They propose a replacement that they claim addresses the major problems, demonstrate a (currently running) implementation, and present an evaluation of it.
The current DNS architecture has long been a blemish on the Internet. In this, I whole-heartedly agree with the authors. As a general rule, any globally required, centrally deployed service is a bad idea in any system where scaling is practically unbounded. The current DNS architecture does fall into this category.
Read the rest of this entry »