mathias
Forum Replies Created
-
AuthorPosts
-
mathiasKeymaster
Dear Atul,rnrn Unfortunately, I cannot help you – I am working with Linux. But it should be possible to run the Makefile. I heart from people working with Eclipse in Windows. You can also run a virtual machine, then you have a Linux system within Windows.rnrnMaybe soneone else of the forum community can share some experience? rnrnMathias
mathiasKeymasterDear Jepson,rnrn A few years ago we were researching this problem in detail (cf. 2009, Heuveline, V. & Krause, M.J. & Latt, J.: “”Towards a Hybrid Parallelization of Lattice Boltzmann Methods””, article in Computers and Mathematics with Applications, 58, 1071-1080). Basically we found that you need to pin the OMP-threads to specific cores to reach almost the performance obtained using MPI. Unfortunately, you need some system calls to do that. Since we want OpenLB to be as generic as possible, we decided against these optimizations. rnrnIn the mentioned paper you will also find some benchmark results to which you can compare yours. However, for a sufficient large grid size I would expect a better speed-up using 48 cores.rnrnMathias rnrn
mathiasKeymasterIndeed there was a bug! We have released a bug-fixed version.rnrnThanks a lot for reporting!rnMathiasrn
mathiasKeymasterHello Klaus,rnrnhere my quick answers:rn1/ You might need to implement a turbulent model which we have done and might put in the next release.rn2/ That depents on the actual problem (geometry, Re-number, inflow-/outflow conditions, etc.) But usually a cluster. I do simulations on 32 to up to several 1000 cores.rn3/ OpenLB does currently not work on GPUs.rn4/ You can easily change the basic data type in OpenLB.rnrnMathiasrn
mathiasKeymasterDear Peijie,rnrnI propose to change the code using the cuboid-parallelisation with “”superLattice””. You will find an example in “”cylinder3d””. The 2d case is very similar.rnrnMathias
mathiasKeymasterDear Peijie,rnrn Can you localise the seg. fault by e.g. using a debuger? In which function call does it happen. Then, send the source code and all input parameter to bug(a)openlb.net. rnrnMathias
mathiasKeymasterYes, I propose to chose a higher resolution of the mesh and smaller uMax.rnMathias
mathiasKeymasterDear Jepson,rnrn It seems to me that the simulation have become unstable after 10000 steps.rnrnMathias
mathiasKeymasterQuote:Quote from mathias on February 19, 2012, 08:37rnQuote:Quote from jepson on February 17, 2012, 02:12rnDear Mathias, rnrnProblems solved ! “”Reynolds number just depends on a characteristic length and speed you defined”” that’s the point, During the simulation, once the reference length and time is determined, Reynolds number is fixed during the simulation, what should to do is just translate the simulation result to dimensionless and then to the physical system. I have read the unit choice of Lattice Boltzmann simulation and refer to your suggestions, that maybe the reason why I confused. rnrnThe Reynolds number is not exactly the physical system Reynolds number only depend on the reference you chose.Am I right now?rn[MATHIAS]: That is right! rnrnmany thanks rnrnJepsonmathiasKeymasterQuote:Quote from jepson on February 17, 2012, 02:12rnDear Mathias, rnrnProblems solved ! “”Reynolds number just depends on a characteristic length and speed you defined”” that’s the point, During the simulation, once the reference length and time is determined, Reynolds number is fixed during the simulation, what should to do is just translate the simulation result to dimensionless and then to the physical system. I have read the unit choice of Lattice Boltzmann simulation and refer to your suggestions, that maybe the reason why I confused. rnrnThe Reynolds number is not exactly the physical system Reynolds number only depend on the reference you chose.Am I right now?rnAnswer: That is right! [Mathias]rnrnmany thanks rnrnJepsonmathiasKeymasterDear Jepson,rnrn You actually fix the viscosity. Then, the Reynolds number just depends on a characteristic length and speed you define. I recommend having a look at “”example/cylinder3d”” to get an idea of how to model a similar situation. If you want, you can send me your problem set-up by email.rnrnMathiasrn
mathiasKeymasterHi Jepson,rnrn A more natural way is to fix the Reynolds number and slowly increase the inflow velocity as done in “”example/cylinder3d”” in the function “”reIniGeometry””.rnrnMathias
mathiasKeymasterDear Jepson,rnrn What are the disrcetisation parameters dx and uMax (cf. log file)? Have you tried to choose a smaller uMax and refine your grid? Is it more stable then?rnrnMathias
mathiasKeymasterDear Pei-Ying,rnrn There will be a new release soon. At the moment there is no puplic git or svn. If you want to contribute just send an email to info(at)openlb.net.rnrnMathias
mathiasKeymasterDear Pei-Ying,rnrn The OpenLB road map strongly depends on the people contributing code. Multi-physics and multi-scales applications in rather complex geometries are topics which will be addressed in the years to come. I guess all your topics are of interest in future. You are welcome to contribute.rnrnMathias
-
AuthorPosts