The maximum size of packet varies from network to network. Several factors will dominate the size of the packets like hardware, the operating system, protocol design, that attempts to reduce the re-transmission overhead in case of errors and to present the packet occupying the channel too long. The obvious solution for forwarding packets when they enter the network having smaller maximum packet size is to fragment the incoming packet and transmit only these fragments. The process of fragmentation is of two types.
When a packet is fragmented upon entering a network ‘A’, and is recombined when fragments leave out of ‘A’, so that the networks are not able to know even that some fragmentation took place somewhere. This is called Transparent Fragmentation.
All the fragments should reach the exit gateway whereas the reassembly takes place. Restricting all the fragments via the same node may cause some performance loss.
Large overhead of repeated fragmentation and reassembly occurs when large packets pass through several small packet networks.
When a packet is fragmented, it is never combined back at intermediate node. The packet is recovered by combining only at the destination node.
Every host should be able to reassemble. In a transparent fragmentation, the network operator can deploy special boxes for this purpose at required points, thus hosts can be freed from this additional overhead.
Because in fragmentation many individual fragments are in the network compared to only on packet, the processing overhead in intermediate routers increase. The overall additional control of information to be added in headers will increase drastically.