From santiagolambrisca at gmail.com Thu Apr 12 20:47:06 2018 From: santiagolambrisca at gmail.com (Santiago Lambrisca) Date: Thu, 12 Apr 2018 15:47:06 -0300 Subject: [SimGrid-user] Simgrid Cluster Speed property configuration Message-ID: Hello! I am trying to compare the times of some apps on real MPI vs SMPI on a few ARM nodes. I would like your help to configure my SimGrid platform. My platform is a Cluster. I would like to know if there is any way to estimate the speed property of my real host. Besides, how the calibrate_flops.py is related to the speed. I got the calibrate_flops.py script from here. > http://simgrid.github.io/SMPI_CourseWare/topic_understanding_performance/ > matrixmultiplication/ > By now, I setted the *speed* value randomly, and setted the *smpi/host-speed* I got from the calibrate_flops.py I compiled both apps (smpi and mpi) with -O3. But then if I fit the speed to reduce error on a cpi example, I get some error on another app doing a simple addition on loop. Can I estimate that speed parameter? or it should be fitted mannually? Thanks in advance, Santiago -------------- next part -------------- An HTML attachment was scrubbed... URL: From cyril.seguin at ens-lyon.fr Wed Apr 25 16:11:11 2018 From: cyril.seguin at ens-lyon.fr (Cyril) Date: Wed, 25 Apr 2018 16:11:11 +0200 Subject: [SimGrid-user] About network topology generator Message-ID: Hello, I'm currently surveying some network topology generators (like BRITE, GT-ITM, ...) that automatically generate different topologies to be used in network experiments. In brief, the purpose is to provide automatic tools to generate networks for SimGrid experiments. I'm interested in the way SimGrid users currently generate their experimental network. By hand? Home-made tools? Other? I'm open to read your feedbacks. Best regards, Cyril. From 869971672 at qq.com Thu Apr 26 17:15:18 2018 From: 869971672 at qq.com (=?gb18030?B?xvDQobXj?=) Date: Thu, 26 Apr 2018 23:15:18 +0800 Subject: [SimGrid-user] Analyze the trace file generated by MPI program Message-ID: I use the MPI simulation module in SimGrid to generate a trace file for the MPI source code simulation run. What does the number in the file represent? The source code and file screenshots are as follows, please help me to answer it, thank you! -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: MPI Program screenshot-1.png Type: application/octet-stream Size: 63336 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: MPI Program screenshot-2.png Type: application/octet-stream Size: 66559 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: MPI Program screenshot-3.png Type: application/octet-stream Size: 19333 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: Trace file screenshot-1.png Type: application/octet-stream Size: 69208 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: Trace file screenshot-2.png Type: application/octet-stream Size: 55423 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: Trace file screenshot-3.png Type: application/octet-stream Size: 53529 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: Trace file screenshot-4.png Type: application/octet-stream Size: 46739 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: Trace file screenshot-5.png Type: application/octet-stream Size: 48044 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: Trace file screenshot-6.png Type: application/octet-stream Size: 16836 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: Trace file screenshot-7.png Type: application/octet-stream Size: 8098 bytes Desc: not available URL: From arnaud.legrand at imag.fr Thu Apr 26 17:24:57 2018 From: arnaud.legrand at imag.fr (Arnaud Legrand) Date: Thu, 26 Apr 2018 17:24:57 +0200 Subject: [SimGrid-user] Analyze the trace file generated by MPI program In-Reply-To: References: Message-ID: Hi, Le 26/04/2018 à 17:15, 起小点 a écrit : > I use the MPI simulation module in SimGrid to generate a trace file for > the MPI source code simulation run. What does the number in the file > represent? The source code and file screenshots are as follows, please > help me to answer it, thank you! Paje traces are auto-described traces (see paje.sourceforge.net/download/publication/lang-paje.pdf). An easy way to extract them is to rely on pj_dump from pajejng (https://github.com/schnorr/pajeng) and R as I've doen here: http://simgrid.gforge.inria.fr/contrib/R_visualization.php I hope it will help you. Come back to us if it doesn't. Best, Arnaud From schnorr at inf.ufrgs.br Thu Apr 26 17:46:12 2018 From: schnorr at inf.ufrgs.br (Lucas M. Schnorr) Date: Thu, 26 Apr 2018 12:46:12 -0300 Subject: [SimGrid-user] Analyze the trace file generated by MPI program In-Reply-To: References: Message-ID: <31c5e94c-292a-b49d-311f-83529ecd9d72@inf.ufrgs.br> Hi, I also recommend using pj_dump (part of pajeng [1]). If you want to understand the CSV-like output of pj_dump, you can check the wiki [2]. A newer version of the document describing the Paje language is available [3]. [1]: https://github.com/schnorr/pajeng [2]: https://github.com/schnorr/pajeng/wiki/pj_dump [3]: https://raw.githubusercontent.com/schnorr/pajeng/master/doc/lang-paje/lang-paje.pdf Best, Lucas On 04/26/2018 12:24 PM, Arnaud Legrand wrote: > Hi, > > Le 26/04/2018 à 17:15, 起小点 a écrit : >> I use the MPI simulation module in SimGrid to generate a trace file for >> the MPI source code simulation run. What does the number in the file >> represent? The source code and file screenshots are as follows, please >> help me to answer it, thank you! > Paje traces are auto-described traces (see > paje.sourceforge.net/download/publication/lang-paje.pdf). An easy way to > extract them is to rely on pj_dump from pajejng > (https://github.com/schnorr/pajeng) and R as I've doen here: > http://simgrid.gforge.inria.fr/contrib/R_visualization.php > > I hope it will help you. Come back to us if it doesn't. > > Best, > > Arnaud > _______________________________________________ > Simgrid-user mailing list > Simgrid-user at lists.gforge.inria.fr > https://lists.gforge.inria.fr/mailman/listinfo/simgrid-user -- Lucas M. Schnorr, UFRGS http://www.inf.ufrgs.br/~schnorr/ From arnaud.legrand at imag.fr Thu Apr 26 18:04:27 2018 From: arnaud.legrand at imag.fr (Arnaud Legrand) Date: Thu, 26 Apr 2018 18:04:27 +0200 Subject: [SimGrid-user] About network topology generator In-Reply-To: References: Message-ID: Hello Cyril, Le 25/04/2018 à 16:11, Cyril a écrit : > I'm currently surveying some network topology generators (like BRITE, > GT-ITM, ...) that automatically generate different topologies to be used > in network experiments. There has been a lot of work on this topic in the years 1995-2005. We used to talk a lot about this when presenting SimGrid. See for example (slides 71-87): https://people.irisa.fr/Martin.Quinson/Teaching/SDR/simgrid-tutorial.pdf Our conclusion back then was probably that the problem is not well formulated and in particular, topology alone is useless without information about the network and CPU characteristics. I don't know if much has been done since. Furthermore, topologies (with cloud, fog, etc.) must have evolved a bit... Anyway, as far as we (Martin, Frédéric, myself, ...) are concerned, most of our efforts rather shifted toward the modeling of real platforms than generating synthetic ones. There are a few uggly perl scripts around allowing to convert brite or gt-itm topologies into SimGrid platforms but I don't think anyone used them since several years so they probably would have to be adapted. If you dig new works on this topic, I'm sure many people on the mailing list would be interested (at least, *I* would be ;) so feel free to share your thoughts and finding on this topic. Btw, do you have any particular use in mind ? Best, Arnaud From cyril.seguin at ens-lyon.fr Mon Apr 30 09:28:58 2018 From: cyril.seguin at ens-lyon.fr (Cyril) Date: Mon, 30 Apr 2018 09:28:58 +0200 Subject: [SimGrid-user] About network topology generator In-Reply-To: References: Message-ID: <008da464-7f0e-e7fa-e431-5cf3e6f7484a@ens-lyon.fr> Hello Arnaud, Thanks for your precise answer. Le 26/04/2018 à 18:04, Arnaud Legrand a écrit : > Hello Cyril, > > Le 25/04/2018 à 16:11, Cyril a écrit : >> I'm currently surveying some network topology generators (like BRITE, >> GT-ITM, ...) that automatically generate different topologies to be used >> in network experiments. > There has been a lot of work on this topic in the years 1995-2005. We > used to talk a lot about this when presenting SimGrid. See for example > (slides 71-87): > https://people.irisa.fr/Martin.Quinson/Teaching/SDR/simgrid-tutorial.pdf > Our conclusion back then was probably that the problem is not well > formulated and in particular, topology alone is useless without > information about the network and CPU characteristics. I agree, topology alone is useless. Some tools allow to specify bandwidth and latency (BRITE, FNSS). It's a good beginning :) > > I don't know if much has been done since. Furthermore, topologies (with > cloud, fog, etc.) must have evolved a bit... Anyway, as far as we > (Martin, Frédéric, myself, ...) are concerned, most of our efforts > rather shifted toward the modeling of real platforms than generating > synthetic ones. There are a few uggly perl scripts around allowing to > convert brite or gt-itm topologies into SimGrid platforms but I don't > think anyone used them since several years so they probably would have > to be adapted. For the moment, the most recent tool I have used date from 2012 (FNSS : A toolchain for simplifying Network Simulation Setup) . The first experiments show it does not scale well compare to others (no more than 6000 nodes for a random flat topology,with p=0.1) but I guess it could be enough. > > If you dig new works on this topic, I'm sure many people on the mailing > list would be interested (at least, *I* would be ;) so feel free to > share your thoughts and finding on this topic. Btw, do you have any > particular use in mind ? Currently, the main idea it to provide tools that allow users to deploy and deploy again the same topology config (nb of nodes, links, latency, bandwidth, ...) to make reproducible experiments. The main application would be "network security" but it can be extended to other application. > > Best, > > Arnaud Best, Cyril.