In packet-based networks information is split up into packets. Each packet contains a part of the information in addition to a header with support information, such as where the packet comes from and where it is going.
The packet-based network then forwards each packet from node to node through the network until it reaches the destination. In literature the network architecture of packet-based networks is frequently divided into seven layers based on the Open Systems Interconnection (OSI) reference model. Functionality of each layer is governed by a protocol that ideally operates independently of the layers above and below.
With the advent of the Internet and its predecessor, the ARPANET28, the ability and need to connect multiple networks together in a seamless way led to the design of the more widely used TCP/IP reference model. The third important reference model, but different from the two previous ones, is the ATM (and B-ISDN ATM). Unfortunately, all models have their shortcoming and none of them can describe the disparity of technologies and protocols used in residential broadband today.
For the purpose of this study, the specifics of the models are less important than the possibility of separating and identifying network functionalities.
This is especially important later in the thesis when analysing how the control of infrastructure at different layers can affect service provision.
For that purpose a slightly modified version of the hybrid reference model of Tanenbaum (1996, p. 44) will be used, as illustrated in Figure 12.
Physical Medium Physical Layer
Data Link Layer Network Layer Transport Layer Application Layer
Data Plane Control Plane
Physical Layer Data Link Layer
Network Layer Transport Layer Application Layer
Control Plane Data Plane
Figure 12, Network architecture reference model
At the bottom of this model is the physical layer that is concerned with transmitting raw bits over a physical medium, which in this thesis is exclusively copper or fibre.
The data link layer takes care of transferring data frames between adjacent network nodes. Examples of data link layer protocols discussed later in
28 The ARPANET was a research network sponsored by the US Department of Defence.
this thesis are Ethernet, Asynchronous Transfer Mode (ATM)29, and Point-to-Point Protocol (PPP). If several links interconnect at a single node, that link can use addresses in the data frames to redirect (switch) frames between links. In computer networks this is referred to as layer 2 switching. In Ethernet, switching can be based on Media Access Control (MAC) address, or Virtual Local Area Network (VLAN) tags. Prioritising or expediting transmission of certain frames based on VLAN tags (or other data frame identifiers) is one way of providing quality management in IP networks.
In contrast to the data link layer that only knows network nodes in its proximity (i.e. takes care of node-to-node transmission), the network layer is responsible for end to end (source to destination) packet delivery. The Internet Protocol (IP) is the most widely used network layer protocol. IP adds a header to all packages that includes unique global addresses of sender and receiver, used to route packets by intermediary nodes. Among other fields, the currently deployed IP version 4 header includes a type of service field. This field can be used to tell routers which kind of service a packet should get or, in other words, prioritise the packet by indicating how important it is compared to other packages. This kind of service differentiation is one way of providing quality management in IP networks but not widely implemented.
The transport layer accepts information from applications and ensures transmission with varying level of reliability depending on the protocol stacks used. When using the IP suite, one of two major transport layer protocols is used, Transmission Control Protocol (TCP) or User Datagram Protocol (UDP)30. TCP is connection-oriented where receiving nodes ensure correctness and send confirmation upon receipt of packages. While this functionality is useful in most web-based services, the delay and inefficiency from functions such as error detection and retransmission make TCP unsuited for real-time transmission. UDP provides an unreliable connectionless datagram service, and is more robust for real-time transmission.
29 As mentioned before, Asynchronous Transfer Mode uses a different reference model. ATM therefore also takes care of functions that are outside the scope of the data link layer.
30 Other less used transport layer protocol exist such as the Stream Control Transmission Protocol (STCP) and Datagram Congestion Control Protocol (DCCP)
In ATM, the ATM Adaptation Layer (AAL) provides network and transport layer services. AAL specifies five types of connections, each with different QoS parameters appropriate for different types of services.
This service classification remains the most appropriate mapping between service requirements and transmission properties, and is described in further detail in section 2.5.3.
2.4.1. Categories of packet-based networks
Regardless of the reference model, layers as well as networks in general can offer reliable or non-reliable services of two types: connection-oriented or connectionless31. In connection-oriented networks a channel is set-up between respective layers of the originating and destination nodes prior to transmission. ATM is a connection-oriented virtual-channel based protocol that additionally can provide guaranteed transmission properties. This is accomplished by having “intelligence”
in each network node (maintaining state) and informing it (reserving resources) along the entire end-to-end path, prior to transmission.
This differs from the method used in pure IP networks where each packet is sent from the originating node without checking if or how the network can get it to the destination node. The edge of the IP networks thereby possesses the intelligence, leaving the IP routers in the core of the network “dumb”, with the simple task of checking the destination IP address against a forwarding table to determine the “next hop”. If the queue for the next hop is long, the datagram may be delayed. If the queue is full or unavailable, an IP router is allowed to drop a datagram.
The result is that IP provides a “best effort” service that is subject to unpredictable delays and data loss.
2.4.2. Quality of Service in packet-based networks
As the transmission delay (known as latency), the transmission delay variance (known as jitter), and the loss of packets increase, the perceived quality of the communication deteriorates. Quantitative measurements of these values are called Quality of Service (QoS) parameters. While advanced coding schemes, such as Forward Error Correction (FEC), can reduce perceived quality decay, successful
31 For simplicity, Tanenbaum (1996; p. 23) uses the telephone system to describe connection oriented service and the postal system to represent a connectionless service.
time multimedia systems are always contingent upon loss and timing constraints with respect to end-to-end QoS requirements.
2.4.3. Managed Networks
The general Internet is segregated into a collection of autonomous systems. An autonomous system is a self-contained set of networks under the same administrative control. It is usually managed by a single corporate entity and therefore also often referred to as a “managed network”. Although connecting to the Internet, the networks are heterogeneous by nature and can build on different underlying technologies and protocols. For this reason different autonomous systems may offer different transmission services such as multicasting, priority treatment to different classes of traffic etc.
In the early days of the Internet it was common for large operators to use their core ATM networks for the underlying transport of IP traffic.
Today, most operators have upgraded or are in the process of upgrading, their core networks based on concepts of Next Generation Networks (NGN) (Traupman et al. 1999). Multi-protocol label switching (MPLS) is a key technology in this process, solving many of the inherent problems of pure IP networks by offering traffic engineering, based on label switched paths (LSP) in a similar manner to virtual channels in ATM (Awduche 1999; Li 1999).
2.4.4. The Internet
The Internet is often referred to as a ‘network of networks’. It consists of numerous autonomous systems (often referred to as managed networks) that interconnect through trunking and peering agreements and obey the same Internet Protocol. Each network node has a unique IP address with which the rest of the nodes can communicate32. If two (or more) communicating nodes belong to the same autonomous system, the transmission may be subject to traffic management policies that guarantee a specific set of QoS parameters. However, the general
32 In reality network nodes can also have private addresses which are not directly accessible through the Internet, using Network Address Translation (NAT) to relay packages through. For simplicity however, I only treat public addresses.
Internet only defines a single transport service class, so when packets transcend autonomous systems they generally lose all QoS guarantees33. The impact of this gap between traffic capabilities and quality guarantees within and between autonomous systems severely affects the competitive situation of service providers on the Internet. Operators of managed networks are increasingly offering advanced multimedia services within their boundaries, using the competitive advantage that their control of local resources offers. However, these upgrades are local to the respective managed network and can therefore provide the operator (and those service providers that are granted access to these features) with a competitive edge in provision of services compared to those offering the same service over the general Internet. However, the effect varies greatly between individual multimedia services. Therefore the next section analyses the nature of different multimedia services and especially their transmission and QoS requirements.