From marbles-isi@mailman.isi.edu Thu Jun 6 21:56:24 2002 From: marbles-isi@mailman.isi.edu (Alejandro Bugacov) Date: Thu, 06 Jun 2002 13:56:24 -0700 Subject: [Marbles-isi] [Fwd: Seminar Announcement] Message-ID: <3CFFCC78.F607F521@isi.edu> I'm not sure whether this is worth a trip to campus but it may be interesting to look at their papers at least to see what they use to model message loss....just an idea....see below: "In particular, we provide analytical models for the Congestion Avoidance and Timeout mechanisms that, under the proper loss probabilities, accurately predict the range of time-scales and the strength of the sustained correlation structure of the traffic sending rate." alejandro. -------- Original Message -------- Subject: Seminar Announcement Date: Thu, 06 Jun 2002 13:37:53 -0700 From: Edith To: leana@usc.edu Does TCP Produce Self-Similar Traffic? Daniel Ratton Figueiredo Department of Computer Science University of Massachusetts at Amherst (Daniel is a student working with Prof. Don Towsley) Friday, June 14th, 11 am SAL 222 The short answer is "No". Contrary to recent claims made in the literature that TCP congestion control creates self-similar traffic, we show here that under realistic conditions, the traffic generated by a long-lived TCP connection, while exhibiting sustained correlations over a predictable and finite range of time-scales, is not consistent with long-range dependence or (asymptotic) self-similarity. Our conclusion is supported by analytical results derived from models of TCP and its internal mechanisms and by a careful statistical analysis of TCP trace data. In particular, we provide analytical models for the Congestion Avoidance and Timeout mechanisms that, under the proper loss probabilities, accurately predict the range of time-scales and the strength of the sustained correlation structure of the traffic sending rate. An existing comprehensive model of TCP is also analyzed and the results show that TCP traffic characteristics are dictated by a combination of the two afore mentioned mechanisms and has a predictable finite range of time-scales under which it presents sustained correlations. Moreover, we analyze the same TCP traces used in a previous studies to "prove" that TCP creates self-similar traffic, but by avoiding a number of pitfalls associated with inferring self-similar characteristics, we arrive at a conclusion that directly contradicts the original claim of self-similarity, but that fully conforms to the behavior predicted by the analytic TCP model. We note that traffic generated by TCP can be misinterpreted and might appear to be self-similar due to the presence of sustained correlations over a range of time-scales. Joint work with A. Feldman, B. Liu, V. Misra, D. Towsley and W. Willinger From marbles-isi@mailman.isi.edu Thu Jun 6 22:04:52 2002 From: marbles-isi@mailman.isi.edu (Martin Frank) Date: Thu, 06 Jun 2002 14:04:52 -0700 Subject: [Marbles-isi] Re: Rehearsal of my thesis defense In-Reply-To: <030801c20d9c$b3c18a60$43800980@MOUNTAIN> Message-ID: <5.1.0.14.2.20020606140406.023917f0@tnt.isi.edu> Argh! This is exactly when Marbles and ATTEND wanted to get together, so that's half the division. Could it be at 3pm? -M At 01:57 PM 6/6/2002 -0700, you wrote: >Hi, > >I'd like to do a rehearsal of my thesis defense tomorrow afternoon: > >-- Date: June 7th, 2002 >-- Time: 2:00pm >-- Location: 10th fl cr > >Please come and let me get your comments. > >Thanks, > > - In-Young > >P.S. I will also send a Meeting Maker message to you. From marbles-isi@mailman.isi.edu Mon Jun 10 19:26:25 2002 From: marbles-isi@mailman.isi.edu (Min Cai) Date: Mon, 10 Jun 2002 11:26:25 -0700 Subject: [Marbles-isi] Checked in na Message-ID: <003301c210ac$54d91f50$b7800980@bugatti> I made some modification to na to force the ExImpExecutive exit when marbles2 solver reaches the time given. - Min From marbles-isi@mailman.isi.edu Mon Jun 10 19:26:38 2002 From: marbles-isi@mailman.isi.edu (Martin Frank) Date: Mon, 10 Jun 2002 11:26:38 -0700 Subject: [Marbles-isi] Re: Checked in na In-Reply-To: <003301c210ac$54d91f50$b7800980@bugatti> Message-ID: <5.1.0.14.2.20020610111739.02e68930@tnt.isi.edu> ok, thanks, will make fresh to test your changes *** all: no more check-ins to na or marble* until Thursday (or earlier if I tell you) please *** Reason is we're heading out to the Marines again, so we may have to do some smallish fixes via CVS. Thanks, yours, Martin From marbles-isi@mailman.isi.edu Fri Jun 14 14:38:36 2002 From: marbles-isi@mailman.isi.edu (Donghan Kim) Date: Fri, 14 Jun 2002 14:38:36 +0100 Subject: [Marbles-isi] code checked in Message-ID: <00cd01c213a8$c8c78e70$1a800980@isi.edu> have update the sa solver residing on ".../marbles2/solvers/centralized/sa" and committed in. if you give the solver run time enough, it's going to return the optimal solution. donghan. ============================================ Donghan Kim Graduate Research Assistant University of Southern California Information Science Institute Distributed Scalable System Division http://www.isi.edu/~donghank Tel: (310) 448-9350 Fax: (310) 823-6714 email: donghank@usc.edu or donghank@isi.edu ============================================ From marbles-isi@mailman.isi.edu Fri Jun 14 22:42:00 2002 From: marbles-isi@mailman.isi.edu (Martin Frank) Date: Fri, 14 Jun 2002 14:42:00 -0700 Subject: [Marbles-isi] Re: two files are there In-Reply-To: <000901c21345$8fd35c60$45800980@voyager> Message-ID: <5.1.0.14.2.20020614143701.0235a008@tnt.isi.edu> At 06:48 PM 6/13/2002 -0700, Donghan Kim wrote: >alejandro, >seems to me that all the bugs are fixed and all the implementations are done >now. > >have attached 2 files, one of which is xls file and the other one is just a >text file which shows its behavior. > >in the "num-filled-graph", i have defined the percentage of number of filled >tasks to be the number of filled tasks found by each solver divided by the >number of possible tasks in the problem. this is reasonable. > >from this initial result, we are able to claim our sa solver is now >OPTIMAL!!! Actually no, I think - it is able to fill all the tasks with resources, but that is probably because there are just plentiful resources. I'm betting that there are either (a) problems for which your solver does not find the optimal-task-value solution, or (b) that your solver is exponentially expensive in the number of tasks. (What're your thoughts?) Cheers, yours, Martin >will try to experiment more cases tomorrow. > >donghan. > >please don't care about the running time of sa solver, i mean, some of the >time has been spent on the printing out the debug messages to trace the sa >so that i can follow up the sa and investigate how it behaves. so haven't >plotted and compared. > >============================================ >Donghan Kim >Graduate Research Assistant >University of Southern California >Information Science Institute >Distributed Scalable Systems Division >http://www.isi.edu/~donghank >Tel: (310) 448-9350 >Fax: (310) 823-6714 >email: donghank@usc.edu or donghank@isi.edu >============================================ > > From marbles-isi@mailman.isi.edu Fri Jun 14 22:50:46 2002 From: marbles-isi@mailman.isi.edu (Alejandro Bugacov) Date: Fri, 14 Jun 2002 14:50:46 -0700 Subject: [Marbles-isi] Re: two files are there References: <5.1.0.14.2.20020614143701.0235a008@tnt.isi.edu> Message-ID: <3D0A6536.3E2ADEDD@isi.edu> Martin Frank wrote: > > At 06:48 PM 6/13/2002 -0700, Donghan Kim wrote: > >alejandro, > >seems to me that all the bugs are fixed and all the implementations are done > >now. > > > >have attached 2 files, one of which is xls file and the other one is just a > >text file which shows its behavior. > > > >in the "num-filled-graph", i have defined the percentage of number of filled > >tasks to be the number of filled tasks found by each solver divided by the > >number of possible tasks in the problem. this is reasonable. > > > >from this initial result, we are able to claim our sa solver is now > >OPTIMAL!!! > > Actually no, I think - it is able to fill all the tasks with resources, but > that is probably because there are just plentiful resources. > > I'm betting that there are either (a) problems for which your solver does > not find the optimal-task-value solution, or (b) that your solver is > exponentially expensive in the number of tasks. > > (What're your thoughts?) > Martin, you are right...I guess what Donghan was trying to say was that in this case it finds the optimal value (copared to some previous version that he had with some problems in the logic). He is not doing a complete search so we cannot prove that found the optimal value. What happens is that in these instances we know what the maximun number of possible tasks and since there are enough resources to find a solution that fills them, then he can say he found the optimal solution. When we go to an overconstrained case (which they are trying to do now...I told them to explore problem with shorter planning horizon to see how this complcated the problem) then we won't be able to say that we found the optimal solution using the SA solver. So the answer to your question is (a). alejandro. > Cheers, yours, Martin > > >will try to experiment more cases tomorrow. > > > >donghan. > > > >please don't care about the running time of sa solver, i mean, some of the > >time has been spent on the printing out the debug messages to trace the sa > >so that i can follow up the sa and investigate how it behaves. so haven't > >plotted and compared. > > > >============================================ > >Donghan Kim > >Graduate Research Assistant > >University of Southern California > >Information Science Institute > >Distributed Scalable Systems Division > >http://www.isi.edu/~donghank > >Tel: (310) 448-9350 > >Fax: (310) 823-6714 > >email: donghank@usc.edu or donghank@isi.edu > >============================================ > > > > > > _______________________________________________ > marbles-isi mailing list > marbles-isi@mailman.isi.edu > http://mailman.isi.edu/mailman/listinfo/marbles-isi -- Alejandro Bugacov bugacov@isi.edu Voice: (310) 448-8269 Fax: (310) 823-6714 http://www.isi.edu/~bugacov From marbles-isi@mailman.isi.edu Thu Jun 20 16:00:55 2002 From: marbles-isi@mailman.isi.edu (Donghan Kim) Date: Thu, 20 Jun 2002 16:00:55 +0100 Subject: [Marbles-isi] new run.sh has been checked in. Message-ID: <002101c2186b$472d8cd0$1a800980@isi.edu> to generate harder problems, have edited "run.sh" under .../marbles2/problemgen/ and checked in. before generating the problems with this shell file, please be careful about the unit (i mean, milliseconds for some, hours for others, and so on) of each argument. you can find the unit of each argument in the code "Gn.mjPackage" under .../marbles2/problemgen. donghan. ============================================ Donghan Kim Graduate Research Assistant University of Southern California Information Science Institute Distributed Scalable System Division http://www.isi.edu/~donghank Tel: (310) 448-9350 Fax: (310) 823-6714 email: donghank@usc.edu or donghank@isi.edu ============================================