Reconsidering Fragmentation and Reassembly The Internet Protocol(IP)'s success for the last 15 years, during which several new technologies emerged, is due to its support for diversity. IP's solution to the use of different maximum packet sizes on different networks is to split packets that are too big into fragments, and to reassemble these fragments at the destination. This project reconsiders several issues related to fragmentation and reassembly IP. Reassembly is reconsidered first. The current reassembly algorithms are too slow. A simple expected case optimization can be used to improve reassembly performance to 38 instructions per fragment if the fragments arrive in FIFO order; a goal for this project is to implement this optimization in the NetBSD UNIX kernel. The project also introduces the new idea of Graceful Intermediate Reassembly (GIR), which is a generalization of the existing IP mechanisms of destination and hop-by-hop reassembly. In GIR, fragments are coalesced at an intermediate router in order to use the largest sized packets on its outgoing interface. It can be shown that GIR always outperforms the usual IP mechanism of hop-by-hop reassembly. A goal for this project is to modify existing IP router code to show that GIR can be implemented economically in routers with small processing and memory costs. The project also reconsiders fragmentation. It is shown that that avoiding fragmentation has costs of its own in terms of increased packet processing and/or round-trip delays. Measurements are described in which TCP performance improves after turning on fragmentation. For example, on Ethernet under NetBSD using a 536 byte segment size for TCP results in a throughput of only about 6Mb/s, whereas a throughput of 8.45Mb/s is obtained using a segment size of 1460 bytes (without fragmentation) and a throughput of 8.82Mb/s using a segment size of 16260 bytes with fragmentation. The project will also investigate simple performance models th at can be used to determine when fragmentation is beneficial. Finally, the project address the major disadvantages of fragmentation. A major problem is that when a fragment is lost, the entire TCP segment must be retransmitted, resulting in reduced or zero goodput under loss. A new mechanism --- dynamic segment sizing, in which the segment size is dynamically reduced after loss --- is introduced, which addresses this problem. Simulations (using a modified NetBSD kernel) will be used to show that dynamic segment sizing keeps the goodput at reasonable levels even under extremely lossy conditions. All mechanisms (reassembly optimization, GIR, and dynamic segment sizing) are orthogonal and can be applied to other protocol suites besides TCP/IP. Ongoing information about the status of this project can be found in http://dworkin.wustl.edu/~varghese/FRAG/fraginfo.html
