
<html><head></head><body bgcolor="#FFFFFF"><div><span>Hi,</span><br><span></span><br><span>On 20.10.2011, at 8.09, abheek saha <<a href="mailto:abheek.saha@gmail.com">abheek.saha@gmail.com</a>> wrote:</span><br><span></span><br>

<blockquote type="cite"><span>Hi,</span><br></blockquote><blockquote type="cite"><span></span><br></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>Does "better" mean better media quality or more bandwidth, or does "better"</span><br>

</blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span>mean "more closely models real-world open-internet performance"?</span><br></blockquote></blockquote>

</blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote>

<blockquote type="cite"><blockquote type="cite"><span>We got smoother goodput and better PSNR, the average media bit rate</span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span>was higher in the case of droptail but the variation in media bit rate</span><br>

</blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span>was also much more.</span><br></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><span></span><br></blockquote></blockquote>

<blockquote type="cite"><span>How did you measure psnr in ns2? We have been struggling with this for</span><br></blockquote><blockquote type="cite"><span>some time now in our own simulations?</span><br></blockquote><blockquote type="cite">

<span></span><br></blockquote><span></span><br><span>The encoder and decoder are external to ns2. They are connected by an interface to the ns2 application. Instead of relying on the system clock the codecs rely on ns2's clock. Basically ns2 polls the codecs at small intervals of 1ms to check if they have data to send. </span><br>

<span></span><br><span>I've mentioned this before: a 300sec simulation can take anything from 30mins to 2hrs. Depending on the complexity of the simulation. </span><br><span></span><br><span>If you are interested this environment has been used here:</span><br>

<span></span><br><span></span>Evaluation of Error Resilience Mechanisms for 3G Conversational Video;<a href="http://www.netlab.tkk.fi/~jo/papers/2008-er-3g-conf-ext.pdf">http://www.netlab.tkk.fi/~jo/papers/2008-er-3g-conf-ext.pdf</a><br>

<span>Rate adaptation for conversational 3G Video;<a href="http://ieeexplore.ieee.org/xpls/abs_all.jsp?tp=&arnumber=5072183&tag=1">http://ieeexplore.ieee.org/xpls/abs_all.jsp?tp=&arnumber=5072183&tag=1</a></span><br>

<br></div><div>There is also Evalvid-RA. <a href="http://www.item.ntnu.no/~arnelie/Evalvid-RA.htm">http://www.item.ntnu.no/~arnelie/Evalvid-RA.htm</a><br><span></span><br><span>If we want to also compare PSNR for the above simulations then I'd prefer to eventually rub the simulations in the real world using dummynet/netem/planetlab. </span><br>

<span></span><br><span></span><br><blockquote type="cite"><span>-best regards,</span><br></blockquote><blockquote type="cite"><span>Abheek</span><br></blockquote><blockquote type="cite"><span>_______________________________________________</span><br>

</blockquote><blockquote type="cite"><span>Rtp-congestion mailing list</span><br></blockquote><blockquote type="cite"><span><a href="mailto:Rtp-congestion@alvestrand.no">Rtp-congestion@alvestrand.no</a></span><br></blockquote>

<blockquote type="cite"><span><a href="http://www.alvestrand.no/mailman/listinfo/rtp-congestion">http://www.alvestrand.no/mailman/listinfo/rtp-congestion</a></span><br></blockquote></div><div><span></span></div></body></html>