Conventional protocol implementation suffers in data-moving operations, limiting the maximum throughput. Experiments have measured the large incidence of these operations in a conventional protocol such as TCP/IP running in a Unix environment. To overcome this drawback protocol implementations have been proposed involving ct processor and network interface that could share the same storage buffer. An analytical model is presented for evaluating the performance of different TCP/IP protocol implementations in end systems. The model uses a closed queueing network, employing a processor sharing service centre for the processing unit and two FCFS service centres for the core memory and network interface. Performance results (throughput, utilisation factor, mean queue length, etc.) allow comparison with conventional, two-copy and single-copy implementations of TCP/IP. Moreover, the model allows estimation of the influence on performance of other design parameters, i.e. memory bandwidth, number of packets in the network, socket buffer dimension and interface data rate.
Performance evaluation of TCP/IP protocol implementations in end systems / Camarda, P.; Pipio, F.; Piscitelli, G.. - In: IEE PROCEEDINGS. COMPUTERS AND DIGITAL TECHNIQUES. - ISSN 1350-2387. - STAMPA. - 146:1(1999), pp. 32-40. [10.1049/ip-cdt:19990246]
Performance evaluation of TCP/IP protocol implementations in end systems
P. Camarda;G. Piscitelli
1999-01-01
Abstract
Conventional protocol implementation suffers in data-moving operations, limiting the maximum throughput. Experiments have measured the large incidence of these operations in a conventional protocol such as TCP/IP running in a Unix environment. To overcome this drawback protocol implementations have been proposed involving ct processor and network interface that could share the same storage buffer. An analytical model is presented for evaluating the performance of different TCP/IP protocol implementations in end systems. The model uses a closed queueing network, employing a processor sharing service centre for the processing unit and two FCFS service centres for the core memory and network interface. Performance results (throughput, utilisation factor, mean queue length, etc.) allow comparison with conventional, two-copy and single-copy implementations of TCP/IP. Moreover, the model allows estimation of the influence on performance of other design parameters, i.e. memory bandwidth, number of packets in the network, socket buffer dimension and interface data rate.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.