In this thesis we compare the energy consumption behavior of three versions of TCP -- Reno, Newreno, and SACK. The experiments were performed on a three hop wireless test-bed, each node equipped with Lucent 11 Mbps DSSS WaveLAN card running FreeBSD-4.3. We measured throughput and energy at the sender node. Our results indicate that, in most cases, using total energy consumed as metric, SACK outperforms Newreno and Reno while Newreno performs better than Reno. The experiments emulated a large set of network conditions including variable round trip times, random loss, bursty loss, and packet reordering.
We also estimated the idealized energy for each of the three implementations (i.e. we subtracted out the energy consumed when the sender was idle) and here, surprisingly, we found that in many instances SACK performed poorly compared to the other two implementations. We conclude that if mobile device has a very low idle power consumption than SACK is not the best implementation to use for bursty or random loss. On the other hand, if the idle power consumption is significant, then SACK is the best choice since it has the lowest overall energy consumption.
Thesis PDF format (798KB), Talk ppt format (1188KB)