Experiment and scenario description. The two experiments carried out in this section aimed at assessing the performance of the Retransmis- sion Control software. The experiments scenario used to assess Retransmission Control is the same as the one used to assess Flow Control, 3. Two machines interconnected by a 1 Gbps Ethernet link were used in the XXXXXX island iLab.t virtual wall aggregate. The NS file used for this setup is: #generated by Netbuild 1.03 set ns [new Simulator] source tb_compat.tcl set m [$ns node] tb-set-node-os $m DEB77-64-STD set n [$ns node] tb-set-node-os $n DEB77-64-STD set link0 [$ns duplex-link $n $m 1000Mb 0ms DropTail] $ns rtproto Static $ns run #netbuild-generated ns file ends. With this setup the physical link was modified so packet losses were possible, using the Linux netem tool to randomly discard packets of the Ethernet interface of node n. The loss rate is a configurable parameter, and setting it to zero gives the maximum throughput possible for the IRATI stack. Given the data already exposed in subsection 3.1, using a CWQL and a credit of 30 SDUs each was enough, resulting in a maximum goodput of 955 Mbps. The other relevant parameters for assessing the re- transmission control implemented in the IRATI stack are the value of the timeout for A-timer and the value of the timeout for the retransmission timer. Both values are linked, and thus they can not be chosen independently. In order to be able to evaluate the costs imposed by the retransmission control software, the experiments were carried out as well with flows using only flow control. The experiments on this section try to assess the performance of the retransmission control in terms of throughput versus loss rate. The experiment scenario used to assess retransmission control is the same as the one used to assess flow control, see Figure 3. In the first experiment, the same value was used for both timeouts, for the A-timer and the retransmis- sion timer, and the effect of the loss rate on the goodput was studied. On the other side, the second experiment studied the influence of Tr value on the throughput obtained when the loss rate was kept constant.
