on IP: IPv4 and IPv6

IP is the main protocol at the network layer of the Internet. Essentially, every data sent by any top level layer i.e. transport and application layer, gets sent as IP datagrams over the Internet. IP datagrams is the building block for internetwork communications provided by IP. IP is meant to be a best effort protocol for sending data over a network, hence it is inherently unreliable. An advantage of this particular design decision is that implementation of IP in network interfaces and routers is relatively simple. Moreover, IP is also connectionless, meaning that IP does not maintain any state information of the datagrams coming its way. Each datagram is handled independently from one another. Datagrams of the same message gets delivered to its destination on possibly many different paths and may arrive at its destination out of order [1]. Hence, a protocol like TCP is needed on top of IP to provide a reliable service needed by most Internet applications.

An IP address identifies uniquely each device i.e. hosts, routers, connected to or in the Internet. IP uses a rather simple and intuitive mechanism in routing datagrams from a source to its destination. Routing is done on a hop-by-hop basis. A routing table is maintained by hosts and routers which they use in forwarding a datagram to the next-hop router or network interface indicated in the routing table entry associated with the datagram’s destination IP address. Using ICMP, a router can build its routing table through advertisement and solicitation messages from other routers [1].

The current widely deployed version of IP, IPv4, uses 32-bit IP addresses amounting to approximately 4.3 billion addresses. With the rapid growth in the deployment of applications, services, hosts, etc. on the Internet, exhaustion of available addresses in IPv4 seems inevitable. As an answer to this likely possibility, the Internet Engineering Task Force developed IPv6, which offers a much larger address space, to succeed IPv4 [3]. IPv6 uses a 128-bit addressing scheme allowing about 2¹²⁸ unique IP addresses. Aside from having a much larger address space and changes in the IP datagram format, other changes were incorporated to IPv6 which include among others: IMCPv6 for automatic host configuration upon connection to a IPv6 network and network level security through mandatory IPSec implementation [2]. Initial deployment of the service has been performed in countries like the USA, CANADA, JAPAN, and CHINA with JAPAN enjoying full government support while CHINA showcased it in the 2008 Beijing Summer Olympics.

References:

[1] Stevens ,W. R. (1993). Internet Protocol. In B. Kernighan (Ed.). TCP/IP Illustrated Volume 1 (). Addison Wesley.

[2] Das, K. IPv6 – The Next Generation Internet. IPv6.com [@http://ipv6.com/articles/general/ipv6-the-next-generation-internet.htm]

[3] IPv6. Wikipedia. [@http://en.wikipedia.org/wiki/IPv6]

on "Congestion Avoidance and Control [2]"

The paper describes a congestion avoidance/control algorithm which has following features:

1. a connection re/starts slow, packet transmission rates starts low and then gradually increases, until such time the connection achieves its state of 'equilibrium'. This prevents the connection from sending big bursts of packets which makes it prone to failure because of constant packet retransmissions.

2. has a 'better' round-trip time variance estimation, which allows it to estimate a more realistic retransmit timeout interval, rto, for succeeding packets. This leads to the variability of the RTT variance used of the rto computation with respect to the medium of communication i.e. satellite links, which leads to increase in performance.

3. when congestion really happens, it employs an exponential retransmit timer backoff, which allows the system to really come into its normal state, no matter what.

4. for congestion avoidance, it uses an increase/decrease algorithm with additive increase and multiplicative decrease components. Unlike in [1] which uses a binary feedback mechanism (incorporated as a bit information in the packet header) in determining the state of the system, their algorithm depends on some assumptions about the inherent properties of "lost packets". That is, lost packets are lost essentially because, the network is congested. So if a connection experiences lost packets, this means that the network is experiencing congestion and it should decrease its load. On the other hand, if the connection continuously receives ACKs, then that means it can try increasing its load. To achieve fairness, the gateway would just have to dropped packets coming from mis-having (abusive) hosts, which in turn would 'trick' the host into believing that the network is experiencing congestion, thus have to decrease its load. Just curious, did this solution work? I still prefer [1], in terms of the feedback mechanism.

Ref:

[1] D.-M. Chiu and R. Jain, "Analysis of the Increase and Decrease Algorithms for Congestion Avoidance in Computer Networks", Computer Networks and ISDN Systems, Vol. 17, 1989, pp. 1-14.

[2] V. Jacobson, "Congestion Avoidance and Control", SIGCOMM '88, Sept. 1988, pp. 314-329.

on the "Analysis of the Increase and Decrease Algorithms for Congestion Avoidance in Computer Networks[1]"

The paper presented a mathematical analysis of increase/decrease algorithms for congestion avoidance in computer networks. Congestion avoidance algorithms allow a network to operate at an optimal level of low delay and high throughput. The authors evaluated the set of increase/decrease algorithms based on the following criteria:

1. the algorithm should allow the communication system to operate at a level of optimal resource utilization (high efficiency).

2. the algorithm not only ensures efficient utilization of shared network resources, but see to it that there is fairness in the allocation of such resources among the users of the system.

3. the algorithm should be distributed to make the tasks of the system and the users simple as possible.

4. the algorithm, starting from an arbitrary initial state, should achieve goal 1 and 2 as fast as possible.

They focused their analysis to a set of increase/decrease algorithms which uses linear controls as control functions. A control function is used by a user of the system in increasing or decreasing its load utilization.

Their analysis used graphical vector representation of the different control combinations to identify the configurations of feasible linear controls that would allow the system to reach the goal of optimal resource utilization and fairness resource allocation as fast as possible. Using this approach, they found out that a simple linear control with an additive increase and multiplicative decrease components is enough for the system to achieve high efficiency and fairness.

Ref:

[1] D.-M. Chiu and R. Jain, "Analysis of the Increase and Decrease Algorithms for Congestion Avoidance in Computer Networks", Computer Networks and ISDN Systems, Vol. 17, 1989, pp. 1-14.

on the 'Rethinking the design of the Internet: 2 The end to end arguments vs. the brave new world'

The author of the paper reiterates the design principles that have been guiding the development of the Internet up to the present, called end-to-end arguments. End to end arguments in the context of the Internet, follow the notion of making the functions of the lower layers of the Internet infrastructure as simple as possible. Any application-specific features should be pulled out of the core infrastructure and should be implemented at the end systems instead. It proceeds by arguing that these design principles have been the key driving factor of the advances and innovations that the Internet has been experiencing since its early days. This position paper was written in the face of increasing interests of third parties i.e. private entities, governments, demanding the inclusion of new features which would allow more “better” mechanisms for providing security, privacy, accountability, etc. The paper cited a situation wherein implementing “eavesdropping” mechanism at the lower level of the infrastructure would still proved useless, after the fact, that end to end points of the communication are free to apply any available mechanism i.e. encryption, etc., to the messages being exchanged. Instead of providing the benefits one expected from it, it would only add complexity to the core network which in turn would increase the cost of deploying new applications to the Internet.

on the 'The Design Philosophy of the DARPA Internet Protocols'

The paper enumerated and described the various goals behind the DARPA Internet Project which gave birth to what we know now as the Internet; at the same time discussed their relations to the mechanisms which were chose to achieve those goals. The Internet started out as a military funded research project under the Defence Advanced Research Projects (DARPA) of the DoD of the USA.

The main goal of its inception was to provide a way for multiplexed internetwork communications among existing heterogeneous and disparate network infrastructures. Packet switching was chose as the technique for multiplexing since most of the existing networks employ packet switches.

Several second level goals were considered in its design, which proved to have great effects to what have become of it now. Survivability which relates to service availability and continuity tops among the second level goals of the design of the Internet. For example, any interruptions in some part of the network should not disrupt the usability of the whole infrastructure. Also interruptions at the lower layers of the infrastructure should be hidden or abstracted from the application level. Support for multiple services came second. So the designers of the Internet wanted to bring as many services to the Internet as possible. If we remember, reliability was the critical design requirement of the TCP. For some services i.e. real-time applications, the reliability of the TCP comes with a cost, performance degradation. The decision to formally define the boundary (layering) between TCP and IP was made, and another transport layer protocol was created, which is the User Datagram Protocol. UDP provides application a low level interface in performing their needs of internetwork communications, resulting to better control, flexibility and performance. Also, they wanted that the Internet will be able to accommodate various types of networks. Other goals which set at the bottom of the Internet priority list where (4) a mechanism for distributed management of its resources should be provided, (5) it must be cost effective, (6) host attachment must be easy, (7) resources in the Internet must be accountable. Interestingly, (7) has not been fully realized until now.

Surprisingly, there was no explicit mention of security in the original design of the Internet. Survivability was there, but I believe its notion relates more on the physical aspect of the infrastructure. The inclusion of the idea of a datagram as a building block element I think is one of the great realizations of the designers of the Internet. It gives developers better control and flexibility in meeting the networking aspects of the applications that they develop. Fate-sharing is another design decision that has proved to be critical in the development of the Internet. The notion of maintaining state information of communications at the end points only, enabled hosts to be less dependent on the performance of intermediary points in the networks. This provides service continuity on cases when some disruption happens at subset of the network.

on the 'A Protocol for Packet Network Intercommunication'

It was 37 years ago when Turing awardees Vinton Cerf and Robert E. Khan published the seminal paper describing TCP (with implicit mention of IP) which eventually led to the development of the Internet.

The paper proposed a protocol which would allow internetwork communications between processes on hosts residing in different packet switching networks. The paper stated the issue of how would such protocol handles communications between existing and planned packet switching network infrastructure, which would likely be different from one another. So it was here that the idea of having a standard protocol, which is as simple and reliable as possible, for inter process and network communications came in. Since reliability is one of the top concerns of the protocol, a mechanism for detection of ‘lost’ packets and their retransmissions was also included. A sender TCP will wait first for the receiver TCP to acknowledge bytes of messages it previously sent. If it does not receive such acknowledgment within a defined timeout, it re-transmits the unacknowledged bytes. State information of the connection between two communicating processes are kept only at both ends of the communication link, thus making the tasks of intermediary points as simple as possible i.e. only handles forwarding of packets and fragmentation if needed. Provisions for flow control was also included which is based on a window strategy. With this flow control mechanism, the receiver TCP will be able to advertise the number of bytes of data (the window) it can handle to the sender TCP, hence controlling the amount of data that flows between the receiver and sender TCP. One should really appreciate the completeness of the mechanisms or features of the protocol suggested in this paper, considering that most of them were given entirely in their pure theoretical sense. Amazing!