Ahmed Ashir

Estimation of network performance: an approach based on network experience

Network application logs (squid logs, ftp logs, mail logs etc.) are a paramount source of network information. By a careful analysis of these logs, a network performance metric (throughput etc.) can be obtained. Since these records contain past information, any conventional analysis would result in network performance indices which are somehow static. On the other hand, network resources as well as the utilities are dynamic. It thus poses a difficult task to understand how the network would behave at some time in future. We present a estimation model of the network behavior dynamics exploiting its past activities. The approached model employs a sophisticated application of standard statistical estimation techniques applied in addition to a variant of genetic algorithm operators. This involves introducing new statistical operators to adapt to their genetic algorithm counterparts. We have carried out experiments using a one year long network application log archive, presented the results and displayed the evaluation of our approach.

An efficient flow control approach for TCP over wireless networks

Recently, mobile and wireless devices are being incorporated to the Internet at a rapid pace. This initiates the need for optimizing major applications to perform satisfactorily over wireless links. TCP performance over wireless networks is affected by high link errors, low bandwidth and large delay. Existing well established solutions like the Eifel algorithm, TCP I&RW, TCP SACK and D-SACK attempt to address these issues. However, Eifel or TCP I&RW suffer from ACK losses, while TCP SACK or D-SACK introduce overhead in addition to failure in the presence of multiple packet loss. In this paper, we propose a scheme that overcomes the above problems by introducing a concept of indirect acknowledgment. We name this scheme as TCP IACK. Our proposal exploits TCP timestamp option as an additional identifier to TCP segments for the purpose. Furthermore, it deploys one padding byte following TCP timestamp option to recover from ACK loss. Our proposal is therefore, lightweight, more efficient and robust than existing approaches. We also verify these performance implications through experiments.