09.30.06
Posted in Paper reviews, network services at 6:36 pm by mbecchi
The proliferation of distributed applications (instant messaging, content sharing, video streaming, routing, system location, etc.) relying on a hashing functionality has fostered the introduction of distributed hash table (DHT) systems whose service can be shared among many different clients. The heterogeneity, competitiveness and lack of trust among the client applications have lead to challenging and partially conflicting requirements. First, the DHT interface must trade off generality and ease-to-use. Second, resource allocation should be fair and flexible, and maximize utilization.
Read the rest of this entry »
Permalink
Posted in Paper reviews, network services at 5:28 pm by Sailesh Kumar
During the last 3-4 decades, gaming industry has grown from a few brave entrepreneurs coming up with hardwired and simple handhelds to several large corporations with multi-billion dollar business. Blockbuster games, today generates as much revenue as blockbuster Hollywood movies. Current generation games feature truly three dimensional and real life graphics, playability and physics. While earlier generation games were usually player by a single player played against a single bot (player simulated by the device), today’s games allow hundreds of players to simultaneously play against each other and against several computer simulated bots.
Read the rest of this entry »
Permalink
09.29.06
Posted in Paper reviews, network services at 9:42 pm by Niarcas
Distributed Hash Tables(DHT) are used in distributed applications such as distributed file systems, peer-to-peer file sharing systems, multicast, and instant messaging. OpenDHT is a free, shared DHT deployment also known as a public storage facility. The designers of OpenDHT proposed that not every DHT-based application needed its own DHT and that an OpenDHT could ameliorate the efforts needed to design, implement, debug and maintain one.
Read the rest of this entry »
Permalink
Posted in Paper reviews, network services at 9:41 pm by Amy Freestone
This paper discusses some rather impressive results for Colyseus, a novel distributed architecture designed for use by multiplayer video games, supporting an order of magnitude more players than existing single server designs while retaining similar bandwidth costs per node. Colyseus is aimed primarily at first-person shooter (FPS) games, which the authors believe to be “the most difficult to support in a distributed setting.”
Read the rest of this entry »
Permalink
Posted in Paper reviews, network services at 9:41 pm by BrandonHeller
Colyseus is a CMU-designed system for distributing the management and update of game state across multiple servers. Rather than use a client-server model, Colyseus uses a form of overlay network to remove the typical client-server issues of robustness and scalability. Other reasons unique to game servers are mentioned; distribution enables higher frame update rates and even creates a new business model for shared game server networks. Given these benefits, plus the trend toward multiplayer games with more players and greater environment interactivity, the motivation for a distributed game server platform is clear.
Read the rest of this entry »
Permalink
Posted in Paper reviews, network services at 9:40 pm by charlie.wiseman
OpenDHT is a public distributed hash table service running on PlanetLab. The biggest selling point is that it is extremely simple for other applications to utilize OpenDHT. Of course, this is the main cause for its primary drawback: limited utility. The authors address this well in the paper, and make a decent argument for their service. Looking past the surface, there are a few problems as well as some debatable statements, but the fact that the service has been deployed and is actually being used by others goes a long way towards showing it is a viable solution in the right situation.
Read the rest of this entry »
Permalink
Posted in Paper reviews, network services at 9:39 pm by harri
This paper describes a distributed system for increasing the scalability of first-person-shooter (FPS) online computer games, using master/slave data replication to spread the load of state update distribution among several game server nodes.
Read the rest of this entry »
Permalink
Posted in Paper reviews, network services at 9:39 pm by nuzhet.atay
The paper proposes a public DHT service that provides a simple to use interface and supports generic DHT-based applications. The authors indicate that one concern about the current DHT-applications is the need for DHT deployment for each type of application. As the variety of DHT applications increase, one question can be whether to continue to follow this design method, or to deploy shared structures to be used among different applications. The authors notice that most applications use common features, so a shared infrastructure can provide the necessary functions for these applications. I think this is a valid concern, considering that many DHT applications are being developed and most of the currently available DHT applications use similar structural components. One framework that satisfy the needs of many applications can both ease application development and decrease network load. For this purpose, the authors propose a framework which is formed of a DHT network that provides a common interface for various applications, and a number of clients running applications on the DHT network using RPC.
Read the rest of this entry »
Permalink
09.25.06
Posted in Essays at 7:11 am by charlie.wiseman
First, let’s get a few possibly jumbled ideas straight. Interfaces sit at the boundaries between separate components of an architecture. For instance, part of the Internet infrastructure is the application programmer’s interface which defines how an application gains access to the lower layers of the network. Indeed, this interface also provides a logical break between the network itself and the entities that use the network. In this context, then, the evolution of networks comes from internal change, while the survivability of networks deals primarily with external change. Although the two ideas are clearly related, the utility of the network is what keeps it alive in the short term, while the ability to adapt to changing conditions drives the evolution in the long term. Given all that, one of the most important factors in enabling networks to evolve and survive is having well-defined interfaces.
Read the rest of this entry »
Permalink
Posted in Essays at 7:10 am by harri
Although layered protocols are seen by some as an unmitigated good, this essay will try to lay bare a fundamental truth: layers doesn’t matter, interfaces do!
To begin with, I offer my definition of what a (strictly) layered design is: a special case of modules communicating via interfaces. Specifically, each module communicates with exactly two other modules and the undirected graph of pair wise communicating modules is acyclic. For our discussion, let the stack of layers be terminated at one end with an application module and at the other by a physical module. We say that a layer A is higher (lower) than a layer B if it is closer to the application module (physical module) than B is.
Read the rest of this entry »
Permalink
09.22.06
Posted in General at 7:46 pm by mbecchi
I have a question about one of the design principles from NewArch report, specifically the one implying the concept of no per-flow state in the network.
I was thinking about the paper we read last week in the networking seminar. In it, there were several examples of possible use of per flow state in routers. One was congestion control for video streaming, one was peer-to-peer traffic detection (even if the methodology was very imprecise) and the other one was traffic measurement. The authors mentioned that these services could also be implemented in other ways (e.g.: congestion control could be implemented through the addition of a priority field in the packet format), but the use of state information would offer more flexibility.
I was wondering whether the “fate-sharing” design principle implies that per flow state should not be used to implement this kind of services on network routers.
Permalink
Posted in Paper reviews, overlay networks at 7:38 pm by Paul Moceri
There is no doubt that overlay networks provide useful, additional functionality to the Internet not present in the native network. However, since the native network is unaware of overlays, performance of overlay networks is fundamentally limited. Due to the rising popularity of overlays, the authors propose that the native network should evolve to provide better support for the requirements of overlay networks. This naturally leads to the question of whether or not such a proposal is a contradiction to the original idea of overlay networks.
Read the rest of this entry »
Permalink
Posted in Paper reviews, overlay networks at 7:37 pm by nuzhet.atay
The paper proposes and discusses ways to optimize usage of overlay networks. The authors claim that overlay networks are providing a good solution for supporting services that native networks cannot support, without needing to make any changes in native layer. The problem with this approach is the limitations imposed on overlay networks by underlying native networks. So, they claim that native networks should be updated in a way to support overlay networks and the services they provide. However, this results in a contradiction in the goals, such that overlay networks are proposed to avoid changing native networks, but achieving good overlay network performance requires changing its underlying native network. The authors are indicating that there are three ways to get necessary support from native layer; adding a new function to native layer, modifying functions at native layer and using function parameters to customize functions to support overlay layer. They claim that the best choice is the third one because only this approach does not require change in the native layer. They name this approach using non-contradictory overlay-friendly native network. Non-contradictory means no change is needed at native layer, friendly means performance of overlay applications is increased without any negative effect on non-overlay or other overlay applications.
Read the rest of this entry »
Permalink
Posted in Paper reviews, CDNs at 7:35 pm by BrandonHeller
CDN’s, or Content Distribution Networks, are one way to get past the availability limitations inherent in a server-client model. In the paper, designers behind the Akamai CDN detail their system for increasing availability, regardless of DDoS attacks, flash crowds, equipment outages, network partitions, and other challenging scenarios. They focus on the ACMS, the Akamai Configuration Management System, which forms the robust back-end to Akamai’s 15,000+ front-end content servers.
Read the rest of this entry »
Permalink
Posted in Paper reviews, CDNs at 7:35 pm by traviskeshav
Note: On 9/26/2006, I made edits on this paper, as I mistakenly believed that ACMS transferred more than simply configuration files. Consequently, I have removed the paragraph concerning the ‘first security concern’, as it is virtually irrelevant, as well as editing sections of the paper where I made incorrect remarks based upon my incorrect assumption.
Content Delivery Networks (CDNs) have become an effective means of distributing both data and media across the world. This paper describes one of the largest commercial CDNs, known as the Akamai Configuration Management System (ACMS). Using over 15,000 servers across 60 countries, ACMS is able to achieve the goals of efficiency, fault-tolerance, availability and reliability, accomplished through the methods described within the remainder of this review. After describing and critiquing its operations, one issue with ACMS will be examined.
Read the rest of this entry »
Permalink
Posted in Paper reviews, overlay networks at 3:44 pm by Niarcas
In this paper the two authors ask if an overlay-friendly native network is a contradiction in terms and assert that it is not. An overlay network exists to obtain new network functionality not provided by the native network. If native networks could provide all the needed functionality for all applications then overlay networks would be useless. In essence, you can think of the overlay network as the frosting on top of a cake. The frosting exists to add taste to the cake as an overlay network exists to add functionality to a native network.
Read the rest of this entry »
Permalink
09.19.06
Posted in Paper reviews, security at 11:13 am by jon.turner
SybilGuard: Defending Against Sybil Attacks via Social Networks, by Haifeng Yu, Michael Kaminsky, Phillip B. Gibbons and Abraham Flaxman.
Presentation by Sailesh Kumar.
Permalink
Posted in Paper reviews, security at 11:11 am by jon.turner
Network Capabilities: The Good, the Bad and the Ugly, by Katerina Argyraki and David R. Cheriton.
Presentation by Paul Moceri.
Permalink
Posted in Paper reviews, network services at 11:09 am by jon.turner
Colyseus: A Distributed Architecture for Online Multiplayer Games, by Ashwin Bharambe, Jeffrey Pang and Srinivasan Seshan.
Presentation by Travis Keshav.
Permalink
Posted in Paper reviews, network services at 11:07 am by jon.turner
OpenDHT: A Public DHT Service and Its Uses. by Sean Rhea, Brighten Godfrey, Brad Karp, John Kubiatowicz, Sylvia Ratnasamy, Scott Shenker, Ion Stoica, and Harlan Yu.
Presentation by Andrew Wan.
Permalink
09.17.06
Posted in Essays at 7:20 pm by Niarcas
From North America to Australia, over one billion people, use the Internet everyday. Many take it for granted and never stop to think about the visionaries who put it together. If it were up to the media, people would think Al Gore invented the Internet, but we all know that’s not true. There is no better example of designing for the future, then the Internet protocol suite. The architects behind the Protocol suite should be all over the news, but many don’t even realize they exist. The Internet, as it stands today, is the achievement of decades of work by scientists all over America. The current protocol suite is an evolution of testing and tweaking of the past protocols. It is a triumph of thoughtful design and engineering, after looking at the missteps of the past. To understand the current protocol suite, we must look at its beginnings. This paper will dive into the history of the Internet protocol suite, the people behind it, and how it all came together to weave a digital fabric that still changes people’s lives even today.
Read the rest of this entry »
Permalink
Posted in Essays at 6:15 pm by traviskeshav
While the Internet has become an omnipresent entity on the Earth, this by no means condones the protocols of the Internet Protocol suite (TCP/IP). The Internet has ‘succeeded’, perhaps, if simply considering its spread as progress. However, this expansion has occurred in spite of TCP/IP, rather than because of it. When examining design decisions, ranging from the earliest choices by naive developers, to the mistakes made with IPv6 in more recent times, TCP/IP is no triumph – it instead provides numerous examples of poor design and expedient choice-making.
Read the rest of this entry »
Permalink
Posted in Paper reviews, net architecture at 9:52 am by mbecchi
One important aspect for network architectures is their ability to evolve and to adapt to new requirements which can emerge over time. Changes can be fostered both by the advent of new technologies and by the necessity to offer and integrate new services. The need for evolvability in the networking context is nowadays a very hot topic. In fact, the Internet, one of the most pervasive network architectures, has demonstrated its ability to accommodate many services and to go through several technology shifts. Nonetheless, it is beginning to show limitations and lack of flexibility.
Read the rest of this entry »
Permalink
Posted in Paper reviews, net management at 9:49 am by harri
Disclaimer: the reviewer is not particularly knowledgeable about networking and has zero practical experience with it (below the socket layer!), but is unafraid to spout off opinions, ill-informed as they may be. You have been warned 
.
The authors present their arguments (in the guise of an “extreme design point”, for intellectual stimulation and debate) for centralizing routing, at least within an AS). They believe this would simplify and enhance network management and allow global optimization of the network for multiple simultaneous objectives. In effect, today’s distributed routing protocols would be used only to form a modest subnetwork for carrying control information such as device queries and configuration commands, while forwarding tables would be updated directly by (a) centralized network management host(s).
Read the rest of this entry »
Permalink
09.15.06
Posted in Paper reviews, net architecture at 7:08 pm by AndrewWan
This is a short Workshop paper on a design methodology the authors believe important in network architecture. Normally a lot focus is put on the principles in a higher level in designing a system change, and the authors complement this approach by analyzing the intrinsic “invariants” of the system.
The short paper has a clear structure in declaring the viewpoints of the authors. I am not totally agreeable with the authors throughout different parts of the paper. In the abstract, the authors claim that “they do not address the question of how to reason about the evolution of the Internet” about the principles used in the first age of Internet. I however believe that those specific principles have been chose to accommodate future evolution.
Read the rest of this entry »
Permalink
Posted in Paper reviews, net management at 7:07 pm by charlie.wiseman
The authors present a high-level framework that they believe should be used in place of the current network structure, primarily as related to the control and management planes. The work is fairly well motived in the paper, although I believe that they tend to stretch the facts a bit. Certainly, most people agree that the something needs to be done to help the management of (large) networks, but their claim that current networks are “fragile” seems a bit of an overstatement. It is true that there are many ways that the overall functionality of the network can be broken, but if networks were really fragile then we would see higher levels of disruption than we do today.
Read the rest of this entry »
Permalink
09.14.06
Posted in Paper reviews, net architecture at 10:07 pm by Michael Roche
The paper describes a new method of designing and analyzing network architectures with the use of what the authors call invariants. By their definition, an invariant is limiting property of the network design. It is what is unable to change when the network architecture evolves because in doing so the network would lose backwards compatibility. The authors discuss that by evaluating these invariants during the design phase, one is able to design a network that is able to evolve for future use.
Read the rest of this entry »
Permalink
09.10.06
Posted in Essays at 6:21 pm by Michael Wilson
Ethernet is unquestionably the dominant LAN technology in the world today. Aside from a few legacy installations, almost nobody uses any other technology in the LAN. Even in 1994, industry estimates indicated that over 40 million Ethernet nodes had been installed worldwide. [1]
The question arises: why has ethernet been so successful?
Read the rest of this entry »
Permalink
Posted in Essays at 6:20 pm by AndrewWan
Click the semi-transparent plastic end into the computer jack, click on the icon on the computer desktop and you are into the wild Web. This’s the Ethernet, which has brought so much change to our life. I quote here from Bob Metcalfe’s blog: “Last year, according to IDC, 33 years after Ethernet’s invention at Xerox Parc, a quarter billion new Ethernet switch ports were shipped worldwide.”
Read the rest of this entry »
Permalink
09.09.06
Posted in Paper reviews, net architecture at 12:42 pm by Amy Freestone
The paper presents three basic tenets which its authors believe should be guiding forces in the future evolution of the Internet. First, future architectures should be designed to allow for change. This means explicitly including mechanisms in the architecture which allow the technology to morph and advance, as without such mechanisms, the infrastructure will become more established in one particular form, making it difficult to change.
Read the rest of this entry »
Permalink
Posted in Paper reviews, net architecture at 12:41 pm by AndrewWan
(My review is an inauthoritative one to a paper by well known elites whose design ideas and visions have epitomized the fundamentals of our Internet. Therefore please allow any laughable comments. Experience matters, so the authors can use “tenet” which I will only be qualified to use “concepts” in my paper.)
The paper revolves around the three tenets, which the authors firmly believed in guiding the evolution of the Internet, particularly in making decisions on Internet network(network of the networks as Professor Turner accentuated in the first class) architecture choices for the future. Interacted among themselves, design for change, controlled transparency and the centrality of the tussle space are the three.
Read the rest of this entry »
Permalink
Posted in Paper reviews, net architecture at 11:44 am by Paul Moceri
Next generation versions of IP promise to solve the many problems that plague the current version of IP used on the Internet. However, little thought is usually given to how such proposed versions of IP will be deployed. The general assumption is that ISPs will naturally move to the next, better version of IP because of demand for the improved functionality. In reality, ISPs have few incentives to move to next generation IP because of economic costs of being an early adopter. Such early adoption threatens to disrupt service of current IP and drive away customers who wish to keep current IP services.
Read the rest of this entry »
Permalink
Posted in Paper reviews, net architecture at 11:44 am by traviskeshav
When reading this paper, at first I felt quite convinced, agreeing with the authors in their goals; methods should be implemented such as to encourage ISP’s to propagate IP architecture development and Internet evolution. After all, in my computer networking courses at Washington University, what do I constantly hear? Detailed and accurate lectures describing the current flaws of the Internet. Constant issues with security, which when given a ‘quick fix’, often only lead to further vulnerabilities. A multitude of theories and papers describing methods of improving Quality of Service or routing, which upon inspection are frequently seen to been unimplementable or are equally flawed as the services they attempt to fix. Perhaps this seems cynical, but I very much agreed with the paper’s comments concerning the limited potential of ‘overlays’ as a long-term solution to the problems with the Internet.
Read the rest of this entry »
Permalink