49 The name space of the DNS is organized hierarchically. Top-level domains (country specific domains such as .de and .uk and generic domains such as .edu, .gov) are rare. They are approved by the Internet Corporation for Assigned Names and Numbers (ICANN) and administered by top-level domain administrators. These administrators in turn grant sub-domains to commercial organizations, academic institutions, government ministries, etc., which in turn receive the authority to grant further sub-domains in their zone. Internet Domain names are written with the most local domain label first and the top-domain label last. The labels of the domain names are separated by a period. The organizations responsible for a domain or a subdomain propagate the information on their domain names and the corresponding IP addresses on authoritative DNS servers. An institution will only be granted a second-level domain if it will operate a name server for this domain according to Internet standards (Comer, 2006: 439). The DNS servers are geographically distributed in the Internet. These servers use specialized protocols to cooperate in mapping domain names to Internet addresses for all domain names in the public Internet. Computers attached to the Internet access a DNS server to translate a URL40 or an E-mail address into an IP address needed for forwarding the data. 3.4.4 Routers and routing The information on the location of a specific network host in the Internet, given by the IP address, is evaluated by specialized computers, called routers. Routers are situated at network nodes and interconnection points where the networks making up the Internet interconnect. These routers have the function of forwarding data packages across network boundaries according to rules specified in so-called routing protocols. When more than one possible path to the final destination is available, routing refers to the process of deciding on a specific route.41 It is necessary to explain the process of routing in more detail here, because the routing functions performed by the different ISPs active in the Internet are the central differentiating characteristics between ISPs to be discussed in the subsequent chapters. The following description of routing differentiates between routing within autonomous systems, where traffic is directed according to the IP address and routing between autonomous systems, where traffic is forwarded using the AS number. Routing within autonomous systems Package-forwarding within autonomous systems follows the principle of “next-hop routing.” A router does not know the complete path to the final destination, but only the next router along the way to the final destination. A router evaluates the destination IP address of any incoming package and sends the package along to the next 40 URL stands for Uniform Resource Locator, a technical term for web-address. 41 More recently the term “forwarding” has become more common than “routing” (Comer, 2006: 93). 50 router on its way to its final destination according to information given in its routing table. A routing table pairs IP addresses (destinations in the network) and routing paths (next router in line to the destination) (Comer, 2006: 97). The information listed in the routing tables of neighboring routers obviously needs to be coordinated, so that the “next-hop” approach leads to packages reaching their destinations without inefficient detours. This coordination can be achieved by network administrators inserting the information listed in routing tables manually. However, because the Internet is dynamic with many hosts being connected and disconnected at any given time, automated systems are considered more reliable and stable compared to manual systems. Such automated systems are based on routing protocols by which routers dynamically learn new routing paths from their neighboring routers. Examples for such protocols are the Routing Information Protocol and the Open Shortest Path First Protocol. Using these protocols, routers advertise their current network capacity and the connections they can establish to neighboring routers and receive analogous information. By this exchange of information, routers acquire a complete picture of network topology within their region and fill their routing tables accordingly. The criteria which determine which route a router should add to its routing table are specified by the routing protocols. Distance-vector routing protocols, such as the Routing Information Protocol, use a hop-count to measure the distance to the final destination, whereby one hop corresponds to one router passed along the way.42 Routers communicating via this protocol advertise which destinations can be reached via their network and the number of hops (the number of routers passed) to the final destination via this particular route. The choice of the path a datagram is to take can then be based on the fixed assumption of shortest path. A router will add that route to its routing table which minimizes the number of hops to the destination. The router increases the hop-count with which the route is propagated to other routers by one before adding this route to its own routing table. Link-state routing protocols, such as the Open Shortest Path First Protocol, allow basing the routing decision on more complex criteria. In addition to sending information on the distance to a particular destination, routers that communicate via this protocol also send information on the state of the connection to the destination. This protocol therefore allows for more complex routing, such as routing according to the type of service (for instance higher quality routes are reserved for real-time applications) or routing in order to balance traffic load (Comer, 2006: 94).43 42 Because traversing networks can take different amounts of time, depending on the technology of the network, the protocol allows for routers to set hop-counts artificially high for connections via slow networks (Comer, 2006: 275). 43 Even though Internet technology, as compared to the technology of PSTN networks, is characterized by the fact that most of the intelligent software is located at the edge of the network, whereas the internal switches generally execute only simple functions, the routing protocols can be used for more refined steering of the Internet traffic. The features of the existing routing protocols, if taken advantage of, can even be used to realize some of the functi- 51 To keep the information stored in routing tables to a minimum, routers initially evaluate only the network prefix of the destination IP address. For each possible network prefix, the routing table stores information on the routing path. In this system “the amount of information a router needs to keep is proportional to the number of networks in the internet, not the number of computers” (Comer, 2006: 35).44 Only when the home network of the destination host is reached, is the host part of the IP address evaluated. The final router in the routing path is connected to the home network of the destination host. It will map a physical address to the IP address of the host and deliver the IP packet directly to the destination host.45 A further simplification of routing tables is that routers, in general, do not explicitly list all network prefixes. Rather, the largest part of the destinations reachable in the Internet will be summed into a default route applicable to all networks for which no explicit routing information is listed in the routing table (Comer, 2006: 100). This default route will point to a router which has a more complete routing table. Figure 3.4 illustrates the logic of next-hop routing. The routing table for router R specifies that all incoming packages with network prefixes 20.0.0.0 or 30.0.0.0 can be delivered directly because router R has a direct connection to both of these networks. For packages destined to networks 10.0.0.0 or 40.0.0.0 the routing table lists the IP address of the next-hop router, to which R has a direct connection. Routing between autonomous systems In the early Internet, routing between networks was very similar to the way routing now functions within an autonomous system. When the Internet grew to dimensions of many thousands of interconnected networks, the routing tables, however, did not scale well to this large number of possible destinations. Autonomous systems were invented as a means of combining networks for the purposes of routing. A group of networks is subsumed using one autonomous systems number such that routing tables need routing information for only this AS number in order to be able to reach all of the included networks. For routing between autonomous systems, one or more routers within the autonomous system collect the information on the networks reachable within the AS and propagate this information (or a subset of the reachable networks) to the rest of the Internet. The protocol by which these routers communicate with the corresponding routers of the other autonomous systems is the Border Gateway Protocol (BGP).46 The information propagated via the BGP protocol is not comparable to the routing information exchanged by routers within autonomous systems. BGP only propagates the fact that a particular network can be reached via the router. When two or more ons that are to be made possible by the IPv6 protocol, for instance the differential treatment of particular traffic. 44 Comer uses the term “internet,” without the capital I, to refer to autonomous systems. 45 The protocol that maps physical addresses to IP addresses is the Address Resolution Protocol (Comer, 2006: 57ff.). 46 The standard common today is the fourth version of the BGP protocol, BGP-4 (Comer, 2006: 254). 52 routers advertise the reachability of this network, a router cannot use this information in selecting an optimal route according to criteria such as number of hops or the quality of the connection (Comer, 2006: 264). Halabi (2000: 101) states correctly, that: “ …the primary difference between intra-AS and inter-AS routing is that intra- AS routing is usually optimized in accordance with the required technical demands, while inter-AS usually reflects political and business relationships between the networks and companies involved.” Figure 3.4: Next-hop routing Source: Comer, 2006: 99 Consistency in routing The routing information propagated in the Internet needs to be consistent and complete in order to avoid routing loops and data loss. The functioning of the Internet therefore hinges to a great extent on the premise that the routes advertised by routers are valid. In the early Internet, a so-called core system of central routers, which kept complete information on all destinations in the Internet, guaranteed this consistency. All routers that were not part of this core system could direct a default route towards Q R S N etwork 1 0.0.0.0. Network 20.0.0.0 Network 30.0.0.0. N etwork 4 0.0.0.0. 20.0.0.5 30.0.0.6 40.0.0.7 10.0. 0.5 20.0.0.6 30.0. 0.7 TO REACH HO STS ON NETWOR K ROUTE TO THIS AD RESS 20.0.0.0 DELIVER DIRECTLY 30.0.0.0 DELIVER DERECTLY 10.0.0.0 20.0.0.5 40.0.0.0 30.0.0.7 (a) An exa mple internet with 4 networks and 3 routers, and (b) the routing table in R. (a) (b) 53 a core router and rely on this router to have the information on how to reach all valid destinations in the Internet. The core system also discarded any data packages with no valid destination and thereby guaranteed that this traffic would not block valuable network capacity.47 Within autonomous systems the consistency of routes is checked in much the same way as with the core system explained above (central routers have information on all possible destinations within the autonomous system). Between autonomous systems, ISPs check that propagated routes are valid by comparing reachability advertisements of other ISPs with information listed by so-called routing registries. These registries contain information on which ISPs have been allocated which IP address blocks (Comer, 2006: 266). Of course the data in these routing registries need to be valid and up-to-date. When the NSFNET was privatized, the NSF awarded a contract for a Routing arbiter project, which was to coordinate the exchange of routing information between the independent commercial operators. Merit networks was awarded this contract and has since played a leading role in Internet routing. Today Merit manages the Routing Assets Database, one of the most popular routing registries used by network operators around the world to register their routes and to send queries on routing problems.48 However, many other routing registries exist in parallel such that there is no central authority among the routing registries. Routing problems can and do occur, when routing registries need time for recognizing and repairing inconsistencies. 3.4.5 Quality of service differentiation in Internet routing software In the beginning of the commercial Internet, Internet access services were generally provided over narrowband local network infrastructure. The bandwidth offered by narrowband communication lines was sufficient to support the standard applications of this time, such as E-mail, file transfer, and remote login. Over time, technological improvements and investments into local telecommunications infrastructure have increased the available bandwidth in the local infrastructure substantially. With the advent of broadband infrastructure, more bandwidth-intensive Internet applications have become available, such as online-gaming, video-on-demand, and voice-over-IP (VoIP). The transition to broadband in Internet access services has also had an impact on the demands on network management and routing functions in Internet backbone services. Historically, routing was programmed as a best-effort service. Packets were treated equally in the backbone network, independent of which applications generated the packet flow. With increased use of bandwidth-intensive applications, 47 On the core system of the original Internet see Comer, 2006: 238ff. 48 Furthermore, Merit conducts projects that are to heighten the accuracy of information in routing registries (Blunk and Karir, 2005). 53 a core router and rely on this router to have the information on how to reach all valid destinations in the Internet. The core system also discarded any data packages with no valid destination and thereby guaranteed that this traffic would not block valuable network capacity.47 Within autonomous systems the consistency of routes is checked in much the same way as with the core system explained above (central routers have information on all possible destinations within the autonomous system). Between autonomous systems, ISPs check that propagated routes are valid by comparing reachability advertisements of other ISPs with information listed by so-called routing registries. These registries contain information on which ISPs have been allocated which IP address blocks (Comer, 2006: 266). Of course the data in these routing registries need to be valid and up-to-date. When the NSFNET was privatized, the NSF awarded a contract for a Routing arbiter project, which was to coordinate the exchange of routing information between the independent commercial operators. Merit networks was awarded this contract and has since played a leading role in Internet routing. Today Merit manages the Routing Assets Database, one of the most popular routing registries used by network operators around the world to register their routes and to send queries on routing problems.48 However, any other routing registries exist in parallel such that there is no central authority among the routing registries. Routing problems can and do occur, when routing registries need time for recognizing and repairing inconsistencies. 3.4.5 Quality of service differentiation in Internet routing software In the beginning of the commercial Internet, Internet access services were generally provided over narrowband local network infrastructure. The bandwidth offered by narrowband communication lines was sufficient to support the standard applications of this ti e, such as E-mail, file transfer, and remote login. Over time, technological improvements and investments into local telecommunications infrastructure have increased the available bandwidth in the local infrastructure substantially. With the advent of broadband infrastructure, more bandwidth-intensive Internet applications have become available, such as online-gaming, video-on-demand, and voice-over-IP (VoIP). The transition to broadband in Internet access services has also had an impact on the demands on network management and routing functions in Internet backbone services. Historically, routing was programmed as a best-effort service. Packets were treated equally in the backbone network, independent of which applications generated the packet flow. With increased use of bandwidth-intensive applications, 7 On core syst m of the original Internet see Comer, 2006: 238ff. 48 Fur hermore, Merit conducts p ojects that are to heighten the accuracy of information in routing registries (Blunk and Karir, 2005).