Project - Noise Calibration Mentor - Stanislav Shalunov Weekly Update 3/10 Developer - Cesar Marcondes Task 1) (June 24th - July 14th) - Working on timing packets inside the kernel at the ethernet layer as well as ip and udp layers ====================================================== README Install Patch http://heterovoip.sourceforge.net/README FreeBSD 5.4 Patch http://heterovoip.sourceforge.net/inst_kernel_FreeBSD_5_4.patch (now augumented with ip_layer timing and udp_layer as well) ====================================================== - On Linux, I'm trying to figure out a way to reduce the I/O impact of logging during noise characterization ====================================================== Using Ramdisk but I'm going to program memory storage directly ====================================================== - I'm running tests using stress tool to verify how it varies the multiple attributes of the system http://weather.ou.edu/~apw/projects/stress/ ====================================================== Results - http://heterovoip.sourceforge.net/initial_stress_results Processing Batch Scripts - http://heterovoip.sourceforge.net/initial_stress_results/process_scripts.tar.gz ====================================================== - Determine UDP dispersion precision using thrulay (looking at deviation to the expected dispersion) ====================================================== Basically, I didn't get before that thrulay uses a poisson process to send UDP, not CBR. The results show the exponential curve of the poisson scheduling at 5msec granularity (100Mbps). But, inside each one of the strong "modes" of the exponential curve it seems that there is some noise around 300 usec and 700usec. And the noise there seems to follow a normal distribution ====================================================== - This week I might work in the evaluation of the sender side to narrow down where does this normal distribution come from. (some more code coming) - I'm going to port the FreeBSD kernel to Linux as well (some more code coming)