- Thursday, March 22, 2007
- An analytical Model for Short_Lived.doc
- Published at:IAJIT
The transmission control protocol (TCP) is widely used in wired and wireless networks. It provides reliable transport services between end-to-end hosts.
Since TCP performance affects the overall network performance, several analytical models were proposed to describe the steady-state throughput of bulk transfer TCP flows (i.e., a flow with large amount of data to send, such as FTP transfers). However, most TCP flows in the Internet world are short-lived to see few losses and they cannot reach the steady-state, consequently their performance is determined by the startup effects such as the connection establishment and the slow start mechanisms. Surprisingly, all of the previous models did not investigate the heterogeneity of wireless networks .The heterogeneity is shown by different characteristics and different segment loss probability for various types of wireless networks such as IEEE 802.11 WLAN and 3G cellular network. Moreover, wireless TCP flows are much shorter than wired flows due to the time varying characteristics of wireless networks.
In this research, a recursive and analytical model is developed and used to determine the performance of TCP in heterogeneous wired-cum-wireless networks in terms of average completion time for the short-lived TCP flows is proposed. The proposed model focuses on heterogeneous wireless networks. The model tackles two different schemes, namely, End-to-End scheme that solve the wireless loss problems by the end hosts and Connection Division scheme that split the connection at base station aiming to distinguish different characteristics between different networks. The proposed model is based on the knowledge of average dropping probability, the average round trip time and the flow size in both wired and wireless links.
The analytical model has been validated by means of simulations and using NS-2 simulator. The performance metric of TCP is the average completion time, which is the time that the source takes to successfully transfer a given amount of data in addition to the connection establishment time. It is shown clearly that the proposed model reflects the behavior of TCP; especially in computing the average completion time that increases as the session size (i.e. the number of data segments in the flow) increases. The simulation results are very much the same as the values obtained from the analytical model. The percentage difference for the average completion time (in Sec) is less than 3% and 6% for the End-to-End scheme and the Connection Division scheme, respectively. The model can be used by engineers to tune some of parameters to see its effect on the behaviour of TCP.