On 3/9/06, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote: > Hi, my comments are bellow ok. > > Quoting Soo-Hyun Choi <[EMAIL PROTECTED]>: > > > On 3/9/06, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote: > >> Thanks a lot Soo-Hyun, I agree that a definitive answer is inexistent, > >> as much as the perfect weights value. I have asked for some researchers > >> which has investigated in bandwidth reservations on DiffServ approach, > >> and only one requested me, bit_rate/10. However, the ns-2 does not > >> accept 0.x values, it gives Floating point exception, so 0.002 is not > >> accepted by the ns-2, only from 1. > > > > Although there is not likely a definitive answer for every RED case > > about the parameters, there is abviously a recommended answer. :-) > > You're completely correct :-), good. > > > > > I am having RED simulations with a floating point q_w value without > > any problem under ns-2.28. So you should not have "Floating Point > > Exception" error by using some floating point values for the queue > > weight setting. > > > > If you have changed "double q_w" to something like "TracedDouble q_w" > > in red.h, you may want to replace it to the original declaration. > > > > The q_w is on its original declaration, and I can assign floating > values successfully, however the values needs to be > 0. At least in my > ns 2.28 the value 0.002 gives "Floating Point Exception". > > Sally advises weights values in order to achieve the best performance > for queue management, as we discussed without QoS consideration. I need > to achieve resource (BW) assurance for flows, and the BW amount > reserved for a certain queue (class) increases as reservations requests > are accepted. > > The CBQ and WFQ general discipline > (http://www.cc.jyu.fi/~sayenko/src/wfq-1.2.4.tar.gz) give functions to > set percentages of links for some queues (e.g. 30% for queue 0), thus > configuring the reservations on demand. However CBQ is too complex and > the WFQ general discipline do not work in DiffServ. > > I am using the WFQ discipline for DiffServ > (http://www.cc.jyu.fi/~sayenko/src/diffserv_wfq.diff), which does not > provide a explicit function for reservations, so I suppose that the > reservation is configured by setting the weights on demand, correct? > The 0.002 weight said by Sally means 2% of the link? >
No, this is wrong. You don't seem to have correct understanding of q_weight on RED. The q_weight parameter on RED is to determine how much portion of the current queue (instantaneous queue) would affect to the computation of the average queue length. let) s_n: the EWA queue length (average queue length) seen by the (n)th packet arrival at the RED queue s_(n-1): the EWA queue length seen by the (n-1)th packet q_n: the instantaneous queue length seen by the (n)th packet q_w: weighting parameter then) s_n = (1-q_w) * s_(n-1) + q_w * q_n So, the resource reservation does NOT related to the q_weight value settings at all. As the above equation, it is just a weighting parameter for average queue length computation. I recommend you to read the original RED paper first (written by Sally) before you play with the RED parameters. Otherwise, it is going to be tricky to convince your simulations to the research community (or your professors) without having correct understanding for all parameters that you're trying to manipulate. Soo-Hyun > It is a little bit complicated :-), but I need to solve this problem > which is giving me a big delay. > > King regards, > > Augusto > > > > > Soo-Hyun > > > > > >> > >> Regards, > >> > >> Augusto > >> > >> > >> > >> Quoting Soo-Hyun Choi <[EMAIL PROTECTED]>: > >> > >> > On 3/9/06, Augusto Neto <[EMAIL PROTECTED]> wrote: > >> >> Hi, > >> >> > >> >> The reasons for configurarion fo the RED parameters given by Mrs. Sally > >> >> Floyd, as could see, takes into account the queue management > >> goal. Decisions > >> >> about the dynamically set of the weights are not clear in the > >> text, at least > >> >> considering QoS. Moreover, it is older than the DiffServ :-) that was > >> >> standardized in 1998. > >> > > >> > That's right. Its goal is to have a proper queue management, and it > >> > did not address about the parameter settings dynamically. However, I > >> > became to understand how I set the parameters dynamically by reading > >> > the article. > >> > > >> >> My problem concerns on choosing the correct value to set the per-queue > >> >> weights according to the bit rate of a flow. For instance, my > >> work consider > >> >> that before the adimition of a flow it is necessary to reserve > >> resources for > >> >> it, its bit rate in the case. So, if the bit rate is 3Mb I need > >> to configure > >> >> the scheduler to reserve 3Mb. In ns-2 the reservation concerns in > >> >> setting > >> >> the weights correct? > >> >> > >> > > >> > RED is exceptionally parameter senstive, and it is hard to say which > >> > values are right for a particular case, as you may know. There are > >> > normally 4 important parameters though. > >> > > >> > 1) max thresh: maxthresh = queue_size / 2 > >> > 2) min thresh: minthresh = maxthresh / 3 > >> > 3) max_p: max_p = 2 * packet_loss_rate > >> > 4) q_w: no definitive answer, but it is almost fine if you leave it to > >> > 0.002 (ns-2 default value) > >> > > >> > This is regarded as a general rule to set the RED paramemters for a > >> > simulation, but not for the real world. > >> > > >> > Soo-Hyun > >> > > >> > > >> >> Regards > >> >> > >> >> Augusto > >> >> > >> >> ----- Original Message ----- > >> >> From: "Soo-Hyun Choi" <[EMAIL PROTECTED]> > >> >> To: "Augusto Neto" <[EMAIL PROTECTED]> > >> >> Sent: Wednesday, March 08, 2006 3:59 PM > >> >> Subject: Re: RED parameters configuration > >> >> > >> >> > >> >> > Hi, > >> >> > > >> >> > I think you'd better read this. > >> >> > http://www.icir.org/floyd/REDparameters.txt > >> >> > > >> >> > Soo-Hyun > >> >> > > >> >> > On 3/8/06, Augusto Neto <[EMAIL PROTECTED]> wrote: > >> >> >> Dear Soo > >> >> >> > >> >> >> I need to choose the correct value for weights of RED queues > >> in DiffServ. > >> >> >> For instance, what is the correct value of the weights to be > >> configured > >> >> >> for > >> >> >> a flow with 1Mb of bit rate? > >> >> >> Regards, > >> >> >> > >> >> >> Augusto > >> >> >> > >> >> >> ----- Original Message ----- > >> >> >> From: "Soo-Hyun Choi" <[EMAIL PROTECTED]> > >> >> >> To: "ben" <[EMAIL PROTECTED]> > >> >> >> Cc: "ns-users" <ns-users@ISI.EDU> > >> >> >> Sent: Tuesday, March 07, 2006 4:00 AM > >> >> >> Subject: Re: [ns] RED queue grows beyond limit on ns-2.28-9 > >> >> >> > >> >> >> > >> >> >> > > >> >> >> > You need to set the queue size by modifying the following two > >> >> >> > lines. > >> >> >> > > >> >> >> > $ns queue-limit $node_(r1) $node_(r2) 100 > >> >> >> > $ns queue-limit $node_(r2) $node_(r1) 100 > >> >> >> > > >> >> >> > > >> >> >> > On 3/7/06, ben <[EMAIL PROTECTED]> wrote: > >> >> >> >> Hi Soo-Hyun Choi, > >> >> >> >> > >> >> >> >> I attached the output of the xgraph in ps format. > >> >> >> >> 1. red.ps - red.tcl run on ns2.27 in fedora core 4 > >> >> >> >> 2. red2.ps - the same red.tcl on ns2.28 in fedora > >> >> >> >> core4 > >> >> >> >> > >> >> >> >> In previous post, I wrote the limit is 100. It is > >> >> >> >> mistake, it is infact limit_ is equals to 25. > >> >> >> >> > >> >> >> >> > >> >> >> >> In the first part the current queue is below 25 > >> >> >> >> packets. However, in the second part, the red queue > >> >> >> >> just grow beyond 25 packets. > >> >> >> >> > >> >> >> >> > >> >> >> >> Cheers, > >> >> >> >> > >> >> >> >> > >> >> >> >> --- Soo-Hyun Choi <[EMAIL PROTECTED]> > >> >> >> >> wrote: > >> >> >> >> > >> >> >> >> > The instantaneous RED queue size shouldn't grow > >> >> >> >> > beyond the actual RED > >> >> >> >> > queue size, and the average RED queue size shouldn't > >> >> >> >> > grow beyond the > >> >> >> >> > maxthresh. > >> >> >> >> > > >> >> >> >> > You may want to check your parameters again. > >> >> >> >> > > >> >> >> >> > On 3/7/06, ben <[EMAIL PROTECTED]> wrote: > >> >> >> >> > > > >> >> >> >> > > Hi ns users, > >> >> >> >> > > > >> >> >> >> > > I some doubt about the RED Queue. I run a > >> >> >> >> > > simple tcl script ( can be download from > >> >> >> >> > > http://nile.wpi.edu/NS/Example/red.tcl) for RED > >> >> >> >> > queue > >> >> >> >> > > with 2 TCP sources. I observed that the RED queue > >> >> >> >> > just > >> >> >> >> > > grow beyond the limit (it is set to 100 packets in > >> >> >> >> > the > >> >> >> >> > > tcl script) when run on ns-2.28, ns-2.28 with > >> >> >> >> > eurane > >> >> >> >> > > extension and ns-2.29 . > >> >> >> >> > > > >> >> >> >> > > It is perfectly ok when run on ns-2.27 and ns-2.26 > >> >> >> >> > > with eurane extension. Any comment on this? > >> >> >> >> > > > >> >> >> >> > > p/s: I'm using fedora core 4 and cygwin on windows > >> >> >> >> > > > >> >> >> >> > > Ok cheers, > >> >> >> >> > > > >> >> >> >> > > Ben > >> >> >> >> > > Master Student. > >> >> >> >> > > > >> >> >> >> > > Send instant messages to your online friends > >> >> >> >> > http://uk.messenger.yahoo.com > >> >> >> >> > > > >> >> >> >> > > > >> >> >> >> > > >> >> >> >> > >> >> >> >> Send instant messages to your online friends > >> >> >> >> http://uk.messenger.yahoo.com > >> >> >> >> > >> >> >> >> > >> >> >> > > >> >> >> > > >> >> >> > >> >> >> > >> >> >> > >> >> > >> >> > >> >> > >> > > >> > >> > >> > >> ---------------------------------------------------------------- > >> This message was sent using IMP, the Internet Messaging Program. > >> > >> > > > > > > ---------------------------------------------------------------- > This message was sent using IMP, the Internet Messaging Program. > >
