Internetworking Concepts

Universal Service: the concept that any two pairs of computers can communicate, overcoming electrical incompatibilities, different technologies and addressing schemes. Increases user productivity.

Internetworking: the use of hardware and software to connect networks together to provide universal service.

Because the layer model (OSI) and the hierarchy model (TCP/IP) keep functions of each layer discrete, users and applications are unaware of physical connections and routes through an internet. This is called a virtual network because the actual physical paths are unknown to users and apps.

Reasons for the likelihood of continued network incompatibilities:

  • Large installed base of differing networks
  • Proprietary interests of corporations
  • Lower costs of computers and networks may mean decentralized decisions, which means lack of a single network within an organization
  • New hardware will create the need for new protocols
  • Persistence of legacy mainframes
  • Failure of standards organizations to coordinate

In this diagram, we have examples of:

  • LAN to LAN communication (lower left)
  • LAN to WAN communication (center left)
  • LAN to WAN to LAN communication (center)
  • WAN to WAN communication (upper right)
And referring again to the diagram above, we can see some layers of the OSI model in action:
 
LAYER HOW IMPLEMENTED IN HARDWARE
1 Repeaters across LANs and WANs copy individual bits
2 Bridges store and forward data link frames between LANs
3 Multiprotocol routers forward packets between dissimilar networks
4 Transport gateways connect byte streams in transport layer
Above 4 Application gateways allow internetworking

Application gateways: devices which can convert between flavors of an application. E.g., an application gateway would convert email using SMTP through Eudora to email in the Outlook format.

Interior/exterior gateway: Gateway is an older name for a router. 

Overview of areas of differences in networks:

TUNNELING: The basic idea here is that when sending a packet across a WAN, and the source and destination LANs are similar, WAN routers will strip the LAN frames (keeping the IP address intact) and package the remainder inside the WAN frame type. The application, transport, and IP headers are bundled up safe and sound inside the WAN frame--it tunnels through the WAN.

FRAGMENTATION (also called SEGMENTATION in ATM networks): Some of you may be wondering: what happens when a large LAN packet (say, an Ethernet or Token Ring frame) hits a small-packet WAN (say, ATM)? Well, there aren't a lot of choices. Basically, the data has to be fragmented, and reassembled before it is delivered to the destination node. 

There are two ways to do this: transparent fragmentation and non-transparent fragmentation. Transparent fragmentation means that as packets traverse an internet, there will be devices along the way which will reassemble the fragments BEFORE the packet(s) hit their destination LAN. Non-transparent fragmentation means that all the fragments will be sent along to the destination LAN, and it will be up to the destination node or a LAN gateway to put the pieces back together. 

Transparent fragmentation has these liabilities:

  • an end-of-packet bit must be added to each packet
  • packets must exit a LAN from the same gateway
  • fragmenting and reassembling creates processing delay
Non-transparent fragmentation has these liabilities:
  • Every host must be capable of reassembly
  • Packet overhead increases, because each packet must have a header

(a) represents transparent fragmentation; (b) represents non-transparent fragmentation

The big problem is keeping the sequencing straight, especially if a packet is lost, a retransmission occurs, and the retransmission takes a different route through different WANs with different packet sizes--so when the original frame is reassembled, the retransmitted packet may be in sequence but may be a smaller size than the other (originally received) packets. 

The solution is for the internetwork protocol to stipulate a base fragment size which will fit through any network (ATM is THE smallest, as far as I know, at 48 bytes [+ 5 bytes overhead]; IP packets are the largest, at 65515 bytes). In this scenario, a packet traversing an internet would have to include fields for:  the original packet number; the number of base fragments in the transportation packet; and a bit to indicate if the last fragment is included in this transportation packet. See diagram below.