Basic Internet Architecture Imagine that both hosts are engaged in separate TCP connections with other, unrelated hosts on the Internet. The host TCP connections originate from addr=192.168.0.12, port=W. and addr=192.168.0.13, port=X. respectively. Because there is no coordination between hosts when they choose their TCP source ports, there is no guarantee that W and Xare different. NAPT remaps both the source port and source address fields to ensure that the individual TCP connections appear unique on the public side of the router. In this example, outbound packets from addr=192.168.0.12, port=W. are modified to appear as packets from addr=128.80.6.200, port=Y. and forwarded to the ISP. Likewise, packets from addr=192.168.0.13, port=X. are modified to appear as packets from addr=128.80.6.200, port=Z , where Z and Y are guaranteed to be different. Packets coming back in from the Internet are modified with the reverse mappings before being forwarded onto the private network. 4.3.2.3 Convenience and Limitations Both NAT and NAPT provide independence from the need to renumber when your public IP address(es) change. Only the NAT-enabled router needs to be aware of any change in the range of public IP addresses assigned to the company s network the hosts remain unchanged. This makes it easy for small companies to change ISPs with minimal disruption of internal network operations. NAPT also allows multiple hosts on a home LAN to access the Internet while avoiding additional charges for more than one IP address. Naturally, all this convenience comes with caveats [RFC2993]. NAT and NAPT break the transparency of TCP and UDP communication between hosts, and require special-case coding to handle other protocols. While hosts on the private network may initiate communication with anyone else on the Internet, the reverse is far more complex. Additional functionality is required in your NAT/NAPT router to enable hosts inside the private network to support well-known servers visible to the rest of the Internet. For example, imagine you have a small corporate site with 200 hosts and three of them want to run publicly accessible web servers. The default http://www.companyname.com web address format actually implies that the web server is listening for HTTP traffic on TCP port 80. However, if you only have one public IP address, the NAPT router can only map inbound dst addr, port=80. traffic to one of your internal hosts, not three. The second and third would-be web servers will either need to give up on their plans, merge with first machine, or configure the NAPT router to utilise nonstandard mappings (for example, mapping ports 8080 and 8081 to the second and third internal machines respectively, and giving external web addresses of the form http://www.companyname.com:8080 and http://www.companyname.com:8081 respectively). Running game servers behind NAT/NAPT is similarly problematic. Many games require the server to register its IP address and port number with a master server (through which potential players find available game servers). But when sitting behind NAT/NAPT, inbound connections (e.g. from new players) are typically only allowed by the NAT/NAPT router if they correspond to a recently initiated outbound connection. But since players initiate contact with the game server, not the other way round, we have a dilemma. (For example, consider Figure 4.17 with a Quake III Arena server running on host 192.168.0.13 at port 27960. Further, assume the NAPT router is
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Inexpensive Web Hosting services
64 Networking and Online Games: Understanding and Engineering Multiplayer Internet Games manually configured to map 128.80.6.200:28000 to 192.168.0.13:27960. The master server sees the Quake III Arena server at 128.80.6.200:28000. However, without special configuration the NAT/NAPT router will not allow new players to actually connect through 128.80.6.200:28000 to the game server itself. We discuss this again in Chapter 12.) NAT/NAPT has its admirers and detractors. Nevertheless, it does serve a purpose for private networks that cannot afford lots of public IP addresses or wish to avoid renumbering of their internal networks on a regular basis. Consumer home routers/gateways invariably support some form of NAT/NAPT functionality. Demand is driven by the deployment of broadband IP access over Asymmetric Digital Subscriber Line (ADSL) or cable modem services, and the fact that many homes have multiple computers. Typically the home router has one Ethernet port to the ADSL modem or cable modem, and one or more Ethernet ports for the internal, home network. (To assist in address management of a small home network, many home routers also support the dynamic host configuration protocol described in the following section.) 4.3.3 Dynamic Host Configuration Protocol The DHCP [RFC2131] automates the configuration of various fundamental parameters hosts need to know before they can become functional members of an IP network. For example, every host minimally needs to know the following: The host s own IP address The subnet mask for the subnet on which it sits The IP address of at least one router to be used as the default route for all traffic destined outside the local subnet. Without these pieces of information, a host cannot properly set the source IP address of its outbound packets, cannot know if it is the destination of inbound unicast packets, and cannot build a basic forwarding table that differentiates between on-link and off-link next hops. DHCP allows hosts to automatically establish the preceding information, and provides two key benefits: The need for manual intervention is minimised when installing and turning on new hosts. IP addresses can be leased for configurable periods of time to temporary hosts. Minimising administrative burdens clearly saves money and time, and increases overall convenience. The benefits of dynamic address leasing become apparent in networks where not all hosts are attached and operational at the same time. 4.3.3.1 Configuring a Host DHCP is a client server protocol. Each host has a DHCP client embedded in it, and the local network has one or more nodes running DHCP servers. DHCP runs on top of UDP, which at first glance suggests a Catch-22 situation with the unconfigured IP
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Inexpensive Web Hosting services
62 Networking and Online Games: Understanding and Engineering Multiplayer Internet Games The router remembers the address mapping it used, so it can reverse the process for inbound packets coming from the ISP. For packets coming back in from the Internet, the reverse steps are as follows: A packet arrives at the NAT-enabled router, with a destination address in the 128.80.6/24 range. The NAT-enabled router looks up the mapping between 128.80.6/24 addresses and internal 192.168.0.*, and replaces the packet s destination address with the private IP address of the intended destination host. Internal routing (within the private network) forwards the modified IP packet to the correct destination. A NAT-enabled router is generally free to use whatever address mapping schemes it chooses. For example, in Figure 4.16 the NAT-enabled router might choose to map internal address 192.168.0.10 to public address 128.80.6.20, or indeed any address in the 128.80.6/24 range. Mappings may be statically assigned, or dynamically generated on-demand. The only requirement is that mappings are unique multiple private host addresses should never map to the same public IP address, and vice versa. 4.3.2.2 Network Address Port Translation A common scenario for home networks is where the ISP (whether regular modem dialup or a broadband service) charges additional monthly fees for a second or third public IP address. The solution is an extended version of NAT called Network Address Port Translation (NAPT) [RFC3022]. An NAPT-enabled router transparently makes multiple hosts on the private network appear to be a single host from the perspective of the public Internet. NAPT extends NAT by additionally manipulating the port numbers of TCP and UDP traffic going in and out of the private network. Consider the scenario of Figure 4.17, where two hosts on a private LAN (192.168.0.12 and 192.168.0.13) are sharing a single public IP address (128.80.6.200). Home subnet 192.168.0/24 192.168.0.13 192.168.0.12 NAPT-enabled router ISP All outbound packets have source address 128.80.6.200 192.168.0.1128.80.6.200 Addresses and ports are re-mapped on the way to ISP Source 192.168.0.12:W becomes Source 128.80.6.200:Y Source 192.168.0.12:X becomes Source 128.80.6.200:Z Figure 4.17 NAPT maps both addresses and TCP/UDP ports to share public IP addresses across multiple private hosts
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Inexpensive Web Hosting services
Basic Internet Architecture external hosts in cases where their network contains a mixture of internal and external connectivity. Interesting problems arise when a previously private network wishes to connect to the Internet. These will be discussed in the following section, along with a currently popular solution NAT. 4.3.2 Network Address Translation NAT is used at the boundaries between IP networks, most often between private networks and the public Internet [RFC3022]. Fundamentally, it solves the problem of a private network whose internal IP address space does not map cleanly into an unused, publicly routable IP address space. 4.3.2.1 Pure NAT Pure NAT is best explained with an example. Consider the situation in Figure 4.16. A company has a private network of 100 hosts, using addresses in the private range 192.168.0.1 to 192.168.0.100. At some point in time, the company wishes for all hosts on its internal IP network to access the Internet. The company contacts an ISP and is allocated a CIDR block of 256 addresses, perhaps 128.80.6/24. If there were only a router between the private network and the ISP, every host would need to be renumbered to a unique address in the 128.80.6/24 range. However, NAT provides an alternative to renumbering. Basically, NAT dynamically modifies the source and destination addresses in packets as they are forwarded between the private network and the ISP. For packets being transmitted out to the Internet, the steps are as follows: Source hosts use their own private address in the packet s source address field. Internal routing forwards packets to the NAT-enabled router linking the private network to the ISP. The NAT-enabled router swaps the source address in each packet with a source address taken from the publicly routable address space 128.80.6/24, and then forwards the packet to the ISP. All outbound packets will have source addresses from Private IP subnet 192.168.0/24 192.168.0.10 192.168.0.12 192.168.0.23 NAT-enabled router ISP 128.80.6/24 public 128.80.6/24 space Figure 4.16 NAT helps map a private address space into the public address space
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Inexpensive Web Hosting services
60 Networking and Online Games: Understanding and Engineering Multiplayer Internet Games Tunnelling over an IPv4 network imposes an additional 20 bytes of overhead (the header of the outer IP packet) [RFC2003]. The MTU of the tunnel s virtual link is 20 bytes smaller than the smallest MTU along the outer packet s path. (RFC 2004 suggests a more efficient encapsulation mechanism incurring only 8 or 12 bytes of overhead but with some loss of generality, for example, fragmented IPv4 packets cannot be tunnelled [RFC2004]. For RFC2004-based tunnelling, the protocol type in the encapsulating/tunnelling header is 55.) When tunnelling over IPv6 networks, the MTU drops by at least 40 bytes (the size of the encapsulating IPv6 header) [RFC2473]. 4.3 Address Management Every IP interface needs an IP address, which raises some very real administrative issues when building big networks. There are two key aspects to address management: Establishing an IP subnet from which you can assign IP addresses Actually assigning individual IP addresses to interfaces. In this section, we will review how address blocks are assigned to customer networks from blocks delegated to Internet Service Providers (ISPs); how NAT can be used to overcome limitations in ISP address assignments; how the DHCP simplifies address assignment to individual devices inside your networks and how the domain name service attempted to decouple the naming of endpoints from the addressing of endpoints. 4.3.1 Address Delegation and Assignment To be part of the wider Internet you cannot pick a subnetwork number and prefix at random. You need IP addresses that are globally unique and routable on the Internet. Such addresses are typically obtained from your ISP, which assigns you addresses from larger blocks allocated to the ISP by regional registries around the world [RFC2050]. For example, ARIN (the American Registry for Internet Numbers) manages space under 204/8 (204.0.0.0/8) and a number of other large blocks of IPv4 address space, and APNIC (the Asia-Pacific Network Information Center) manages space under 218/8 and a number of other large blocks. An ISP who asks for space from APNIC will receive an allocation under 218/8 or one of APNIC s other address blocks. Regional registries develop their own policies for subdividing the address blocks they manage. Up-to-date information on registries and assignment policies can be found in the Internet Assigned Numbers Authority web site, http://www.iana.org. Problems arise when you decide to change ISPs. You will usually be forced to adopt a new IP address space assigned by your new ISP (called renumbering). Renumbering of your network is usually required by ISPs because routing would become more convoluted if the address hierarchy was allowed to arbitrarily diverge from the hierarchy of actual connectivity among the service providers. A number of IPv4 address blocks (10/8, 172.16/12 and 192.168/16) have been reserved for use in private internets [RFC1918]. These are useful when building IP networks that will never be connected to the Internet, or will be connected only in a very limited fashion. In principle, such IP networks could be built using any prefixes. However, using designated private IP address spaces helps administrators distinguish between internal and
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Inexpensive Web Hosting services
Basic Internet Architecture For the curious reader: Under many versions of Unix (for example, FreeBSD or Linux) the command ifconfig shows a variety of details about a host s currently attached link layers, including the currently assigned IP address(es), subnet mask(s), and MTU(s). If you are running Windows XP or something similar, the command ipconfig /all at a console window will also print out a variety of details about the host s current configuration. 4.2.3.5 First and Last Hops Forwarding tables are not just for routers. Hosts also have a limited forwarding table that tells them the initial next hop (often referred to as the first hop) for outbound packets. The first hop will either be to a router (for traffic destined beyond the local subnet) or directly to a neighbour on the same subnet. In cases where the next hop goes directly to a neighbour on the same subnet it is referred to as the last hop. (The first and last hops may be one and the same in the case of communication between two hosts on the same subnet.) Forwarding tables have special rules for subnets that are directly attached to one of the host s or router s IP interfaces. Rather than having a specific entry for every IP interface reachable on a local subnet, the next-hop IP address is copied directly from the packet s destination address field. The ARP cache is then scanned for a match to this next-hop IP address, and the packet transmitted to the link layer destination found in the ARP cache. Hosts typically only have a few entries in their forwarding table, for example, one entry for the directly attached subnet, and a default route (network/prefix 0.0.0.0/0) pointing to a router on the local subnet that provides access to the rest of the network. If a host has link interfaces to multiple IP subnets, it will have forwarding table entries for each directly attached subnet, and possibly multiple forwarding entries for nonlocal traffic. For the curious reader: If you are running Windows XP or something similar, entering route print in a console window will show the current forwarding rules. Under many versions of Unix (for example, FreeBSD or Linux) the command netstat -rn will show the host s current forwarding rules. 4.2.3.6 Tunnels as Links Links are simply mechanisms for getting an IP packet from one router to another. A link may even be an IP network in its own right. Transmission of IP packets within other IP packets is known as IP tunnelling, and the link is known as an IP tunnel.From the perspective of the outer IP packet, the packet being tunnelled is just another payload (as uninteresting as a TCP or a UDP frame). From the perspective of the tunnelled packet, the tunnel looks like another link layer. From an implementation perspective, an IP tunnel is a link layer where source and destination addresses also happen to be IP addresses. The tunnel s endpoint is the IP interface identified by the destination IP address in the outer packet s header. When the outer IP packet reaches its destination, the original (inner) IP packet is extracted and processed as though it had arrived over a regular interface. The outer packet s IP Protocol Type identifies the payload as a tunnelled packet, for example, protocol type 4 indicates that the payload is an IPv4 packet [RFC2003]. Because a tunnel represents a single hop from the perspective of the tunnelled packet, its TTL is decremented by one (rather than the number of hops between the tunnel endpoints).
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Inexpensive Web Hosting services
58 Networking and Online Games: Understanding and Engineering Multiplayer Internet Games For the curious reader: If you are running Windows XP or similar, entering arp -a in a console window will show the current local ARP cache entries. Under many versions of Unix (for example, FreeBSD or Linux) the command arp -an will show the current local ARP cache entries. 4.2.3.3 Time to Live Transient errors in router forwarding tables can sometimes create loops in the IP network, known as routing loops. Routing loops tend to occur shortly after topology changes, while the network s routing protocol converges on a new set of shortest-path trees. Routing loops often act like black holes in the network packets head into the region of the routing loop, and then get stuck, consuming bandwidth as they circulate. Under extreme circumstances, the routing loop can disrupt the routing protocol itself, by saturating links carrying routing protocol update messages. To prevent endless looping, IP packets carry an 8-bit Time to Live (TTL) field (see Figure 4.5). In practice, the TTL represents hops to live a limit on the maximum number of router hops a packet can traverse before it expires in transit. A packet s TTL field is set to a nonzero value by the source, and is decremented by one every time the packet passes through a router. The packet is discarded when its TTL field is decremented to zero (whether or not it has reached its final destination). If the source sets the TTL too low, some distant regions of the Internet may become unreachable. (Indeed, there were examples of this occurring with a popular PC operating system s default TTL in the early 1990s as the Internet became more topologically convoluted.) Setting the TTL too high increases the potential disruption a source s packets can cause during routing loops (by increasing the length of time the packets stay in transit around the loop). Many operating systems today set their initial TTL to 64 or some multiple of 32 above that (more by historical quirk than any particular mandatory requirement). 4.2.3.4 Maximum Transmission Units and IP Fragmentation Although in principle, IP packets may be as large as 64 K bytes, most link layer technologies impose a substantially smaller limit on link level frame size. For example, Ethernet imposes a limit of 1500 bytes on the size of IP packets that can be carried in an Ethernet frame. The underlying link layer s frame size limit is reflected at the IP layer by a parameter known as the Maximum Transmission Unit (MTU). When forwarding an IP packet larger than the link s MTU, IP interfaces must perform IP fragmentation chopping (fragmenting) the IP packet up into a sequence of smaller IP packets that all fit under the MTU limit. IP fragmentation occurs underneath the TCP or UDP layer, which allows the source UDP-or TCP-based applications to be unaware of the actual MTUs of links along the path to the destination. The ultimate destination is responsible for reassembling the fragments into the original packet, and then treating the reassembled IP packet as though it had arrived in one piece. IP fragmentation tends to occur when a packet s path originates on a link with a large MTU, and then at some point along the route passes across a link with a smaller MTU. It is not considered a good thing, as it creates less efficient data transfer along the path [RFC1191][RFC1981].
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Clan Web Hosting services
Basic Internet Architecture Link layer IP layer Interface 1 Interface 2 Interface 3 ARP Link layer IP layer ARP Link layer IP layer ARP Link layernetwork Figure 4.14 ARP is both peer of, and service for, the IP layer. Both sit over the Link Layer Ethernet frame Interface 1 Interface 2 Destination Source MAC.1 ARP Request Where is 136.80.1.5? EtherType 0 806 Destination MAC.1 Source MAC.2 ARP Reply 136.80.1.5 is at MAC.2 EtherType 0 806 Destination MAC.2 Source MAC.1 IP packet destined for 136.80.1.5 EtherType 0 800 TimeTime Broadcast Unicast Figure 4.15 Frame sequence when initially sending an IP packet from Interface 1 to Interface 2 is ARP for Ethernet [RFC826], usually just referred to as ARP. When an IP address cannot be located in the local interface s ARP table, the interface issues a broadcast ARP Request on the LAN, essentially asking anyone else if they know the mapping. (Transmitting the ARP Request to the Ethernet broadcast address ensures that all attached devices are reached without the local interface needing to know who is, or is not, attached to the LAN at any given time.) Usually, the target (the interface whose IP address is being queried) will respond with a unicast ARP Reply containing the requested IP to Ethernet address mapping. Figure 4.15 shows the packet exchange that would have occurred in the example of Figure 4.12 if Interface 1 s local ARP table did not have a mapping for 136.80.1.5. (An EtherType 0×806 indicates an ARP Request or Reply, while EtherType of 0 800 indicates an IPv4 packet.) The sequence would be as follows: 1. Interface 1 transmits a broadcast ARP Request for 136.80.1.5. 2. Interface 2 unicasts back an ARP Reply (it knows the Ethernet address of Interface 1 from the initial ARP Request) 3. Interface 1 unicasts the IP packet to Interface 2.
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Clan Web Hosting services
56 Networking and Online Games: Understanding and Engineering Multiplayer Internet Games Destination 136.80.1.5 Source w.x.y.z TCP or UDP payload Destination MAC.2 Source MAC.1 IP packet IP packet Ethernet frame EtherType 0×800 Figure 4.13 IP packets from the router at 136.80.1.2 to the host at 136.80.1.5 are encapsulated for transmission between Ethernet addresses MAC.1 and MAC.2 For example, if Interface 1 belonged to a router whose forwarding rules had just decided send this packet to the interface identified as 136.80.1.5 , the following two steps would be executed A mapping would be established from 136.80.1.5 to MAC.2, the link layer address of Interface 2. The IP packet would be sent from MAC.1 to MAC.2 inside a suitably constructed Ethernet frame. Figure 4.13 shows a highly simplified picture of how an IP packet from outside the 136.186.1/24 subnet would be encapsulated inside an Ethernet frame to be sent from Interface 1 to Interface 2. Ethernet frames carry an ethernet protocol type code EtherType of 0×800 to identify the payload as an IP packet (or more precisely, an IPv4 packet). 4.2.3.2 Address Resolution Next-hop IP addresses are mapped to link layer addresses in a process referred to as address resolution. Address resolution may occur using information that is manually configured or is dynamically discover on-demand. Manual configuration is unwieldy in all but the simplest of static network configurations. Most routers and hosts implement a dynamic Address Resolution Protocol (ARP) to identify what link layer address is associated with a particular IP address. ARP allows IP interfaces to move from one link layer interface to another without manually reconfiguring all the other interfaces attached to the link. This can be useful when, for example, an Ethernet card is replaced on a host or router the Ethernet address changes, and the dynamic ARP process will ensure other interfaces on the link soon learn the new mapping. Figure 4.14 attempts to capture how ARP is both a peer of, and a service for, the IP layer. Interfaces keep current address mappings in a local cache, an ARP table,which is searched each time a packet is transmitted. If a mapping exists, it is used. Otherwise the ARP is executed to discover the desired mapping. To ensure that old or incorrect mappings are regularly refreshed, cached ARP table entries are deleted after a period of time. Each link layer technology has its own ARP mechanism. Some examples are ARP for IP over FDDI [RFC1390] and IP over ATM [RFC2225]. Perhaps the longest serving example
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Clan Web Hosting services
Basic Internet Architecture that have no agreement to directly exchange traffic. Two geographically close sites (for example, London and Paris, Sydney and Melbourne, or Los Angeles and San Francisco) might find themselves communicating over paths that loop through New York, Tokyo, or Amsterdam depending on their choice of backbone provider and where the backbone providers interconnect. 4.2.3 Per-hop Packet Transport This section reviews how an individual IP interface (whether on a router or an endpoint) uses the services of an underlying link to get IP packets to the appropriate next hop. 4.2.3.1 Link Layer Networks For most of this chapter links have been treated as simple, point-to-point paths with only two interfaces attached. In reality, links are often networks in their own right. LAN (such as Ethernet) and wide area networks (such as frame relay) are examples of the link layers that support multiple attached devices. Devices attached to a link layer network may support IP, some other services, or a mixture of both. Two addresses are associated with an IP interface attached to a given link layer network: The interface s IP address (representing the interface s identity in the IP topology). The interface s link layer address (representing the interface s specific identity in the context of the underlying link layer network). In general, the link layer network is unaware of the IP address assigned to any given IP interface attached to the link. An IP packet s next hop (expressed as an IP address in a router s forwarding rules) must be translated to a link layer address before packet transmission can occur across the link. Consider Figure 4.12, where a single Ethernet LAN [8023] has three attached interfaces, belonging either to routers or hosts on the 136.80.1/24 subnet. At the IP level, Interface 1 is known as 136.80.1.2, Interface 2 is known as 136.80.1.5, and Interface 3 is known as 136.80.1.9. Yet the Ethernet LAN only knows these interfaces by their 48-bit (6 byte) Media Access Control (MAC) addresses (in this example MAC.1, MAC.2, and MAC.3, respectively). MAC.1 136.80.1.2 MAC.3 136.80.1.9 MAC.2 136.80.1.5 Interface 1 Interface 2 Interface 3 Ethernet LAN carrying IP subnet 136.80.1/24 Figure 4.12 Each interface on an Ethernet LAN has both IP and Ethernet addresses
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Mac Web Hosting services