WEBVTT 00:00.440 --> 00:01.340 Hello everyone. 00:01.340 --> 00:05.000 Welcome back to the CFD using Openfoam beginner to intermediate course. 00:06.110 --> 00:07.610 This is our class six. 00:07.610 --> 00:13.520 In this class we will be seeing introduction to Transient solvers decomposing methods which will be 00:13.520 --> 00:15.920 useful in running cases in parallel. 00:16.520 --> 00:22.280 If you are interested in running parallel processing, then we will also see how to run parallel processing 00:22.280 --> 00:23.570 for snappy hex mesh. 00:23.570 --> 00:27.800 So the entire meshing process through snappy hex mesh can be done in parallel. 00:27.800 --> 00:29.960 So we will see how to do that. 00:30.680 --> 00:31.040 Okay. 00:31.040 --> 00:33.170 Now introduction to transient solvers. 00:33.200 --> 00:37.550 So transient solvers it uses time dependent simulation. 00:37.550 --> 00:43.520 So we will be getting a sort of animation where every time step is converged solution. 00:43.520 --> 00:48.350 So we will get a feel of the flow progressing over time. 00:49.580 --> 00:56.360 The key features are the model here is unsteady or transient flows such as vortex shedding behind a 00:56.360 --> 00:56.840 cylinder. 00:56.990 --> 01:02.600 Example solvers in Openfoam are pimple foam, eco foam, piezo foam, inter foam. 01:04.010 --> 01:07.850 So applications are like uh, in most cases. 01:07.850 --> 01:13.340 The examples are aerodynamics, combustion, or multiphase phenomena like mixing processes. 01:13.370 --> 01:17.860 The advantage of using a transient solver is to capture time dependent phenomena. 01:17.860 --> 01:19.570 Analyze dynamic behavior. 01:19.660 --> 01:25.300 So in solver settings, apart from what we have done till now, we will also define our time step which 01:25.300 --> 01:26.080 is delta t. 01:27.100 --> 01:32.680 And we will be seeing how to set the convergence criteria which defines solution accuracy. 01:32.710 --> 01:36.220 We will also set maximum Cuda number, serial number and so on. 01:36.910 --> 01:40.960 So we will be seeing a multiple tutorial cases. 01:41.110 --> 01:44.680 So the first one will be backward facing step 2D case. 01:44.680 --> 01:47.230 In openfoam terms it's called pitch daily. 01:48.040 --> 01:50.710 Uh we will be seeing k epsilon turbulence model. 01:50.710 --> 01:53.470 And the solver which we will be using is pimple four. 01:54.310 --> 02:00.880 So as you can see this is the part of control dict file which we will control the entire simulation 02:01.750 --> 02:02.770 parameters. 02:02.770 --> 02:06.430 So the start frame uh start from is set to latest time. 02:06.430 --> 02:12.340 So basically when we don't have any time steps before running the simulation, zero is the latest time. 02:12.340 --> 02:13.900 So it will start at latest time. 02:13.900 --> 02:20.890 But if you want to start at start time every time, then you can start, uh, you can set the start 02:20.890 --> 02:23.440 from to start time instead of latest time. 02:23.710 --> 02:26.890 Too many times in a sentence stop. 02:26.890 --> 02:29.200 It will be n time and you can set the end time. 02:29.200 --> 02:35.530 That is the uh, time flow time in seconds where your simulation will end. 02:35.530 --> 02:39.010 In this case, your simulation will end at 0.3 seconds. 02:39.130 --> 02:42.130 So it's like you're starting a simulation, you're starting a flow. 02:42.130 --> 02:49.570 And after, if you stop the simulation or stop recording the simulation after 0.3 seconds, whatever 02:49.570 --> 02:52.960 video you would have got, that is exactly what we are trying to simulate here. 02:53.050 --> 02:55.690 Then we have the delta T, which is the time step. 02:55.690 --> 02:59.320 In this case it is one over one E minus four. 02:59.590 --> 03:01.780 So that is very small. 03:01.780 --> 03:04.210 And we will see why we said that okay. 03:04.210 --> 03:10.030 In the control deck we are sorry in the right control we are setting it to adjustable and the right 03:10.030 --> 03:21.960 interval is 0.01 which means our delta is 0.0001, but for every 0.01 seconds, we will be saving a 03:21.960 --> 03:26.220 time step file which will cater us to the animation finally. 03:26.220 --> 03:31.950 So we will get a frame of reference for every 0.01 seconds. 03:32.700 --> 03:36.240 So the purge rate is mostly useful only for steady state case. 03:36.240 --> 03:43.050 What it means is if I said purge rate to five, then only the latest five time steps will be stored. 03:43.050 --> 03:46.170 Everything else will be deleted to save some storage space. 03:46.740 --> 03:48.630 Uh, the right format should be Ascii. 03:48.930 --> 03:51.990 You don't have to touch any of these if you just come. 03:52.170 --> 03:52.320 Uh. 03:52.320 --> 03:54.690 Below you will find runtime modifiable. 03:54.690 --> 04:01.260 You can set it to S, because if you see that the simulation is running and you feel like 0.3 is not 04:01.260 --> 04:07.380 enough, then you can, uh, set the end time to 0.5 while the simulation is running. 04:07.410 --> 04:10.980 If you want to do that, then you have to set the runtime modifiable to S. 04:11.580 --> 04:18.240 And in the adjust time step, if you keep s, then what it means is you are doing adaptive time stepping. 04:18.240 --> 04:24.150 So in case of adaptive time stepping you also have to define your maximum current number. 04:24.150 --> 04:26.640 Only then it will uh do it. 04:26.640 --> 04:29.610 So generally maximum current number has to be one. 04:29.610 --> 04:31.410 That is the realistic case. 04:31.890 --> 04:36.810 Uh, it is also popularly known as CFL current Frederic uh Levy number. 04:36.810 --> 04:40.290 But in open form terms it's maximum current number. 04:40.290 --> 04:43.890 We can monitor current number while the simulation is running for every time step. 04:43.920 --> 04:47.400 So by default this tutorial comes with a max of five. 04:47.400 --> 04:49.170 So let's stick on to the tutorial. 04:49.170 --> 04:55.140 But when you are actually setting up a case, make sure that mostly it is one or depends on the physics. 04:56.100 --> 05:00.360 So the next image is from turbulence properties file. 05:00.360 --> 05:06.300 The simulation type is going to be Arias and the Arias model is going to be epsilon k epsilon. 05:06.600 --> 05:10.860 The turbulence is turned on and the print coefficients also turned on. 05:10.860 --> 05:16.200 What this does is in the command prompt or the console window, you will be able to see the coefficients 05:16.200 --> 05:20.660 of turbulence parameters like k, epsilon and duty while the simulation is running. 05:21.710 --> 05:31.490 So after running the simulation, we will be getting, uh, multiple files from 0.010.020.03, so on 05:31.490 --> 05:33.020 to 0.3 seconds. 05:33.020 --> 05:40.250 We will be visualizing those things, those, uh, frames of animation in Paraview after running this 05:40.250 --> 05:40.580 case. 05:40.580 --> 05:43.370 So this is what we are trying to, uh, visualize. 05:43.370 --> 05:48.560 So the left one is 0.01 seconds and the right one is 0.3 seconds. 05:48.560 --> 05:49.550 This is how it starts. 05:49.550 --> 05:50.660 And it ends here. 05:50.660 --> 05:52.670 So this is just a backward facing step. 05:52.700 --> 05:56.840 Now we will jump into our command prompt and try to simulate this case. 05:57.650 --> 05:59.780 I am in class six folder. 05:59.780 --> 06:04.130 Here we have the first pitch daily tutorial. 06:04.400 --> 06:07.250 Here we have zero constant and system. 06:07.250 --> 06:11.660 So as you might already know, uh, pitch daily comes with a block mesh. 06:11.750 --> 06:14.630 So I will run block mesh okay. 06:14.930 --> 06:17.420 And we will find out. 06:17.420 --> 06:19.130 What is this, uh, solver? 06:19.130 --> 06:22.370 As we already saw, this is a simple form case. 06:22.370 --> 06:28.850 Sorry, this is a simple form case, so I will write it to simple form. 06:30.560 --> 06:33.560 And let's wait for the simulation to end. 06:33.590 --> 06:35.930 Meanwhile, I'll just pause somewhere. 06:35.930 --> 06:41.720 Like if you scroll your mouse over here, it will snap to specific time and it will keep on running. 06:41.720 --> 06:45.590 It won't stop, but you can just scroll and it will just stop there. 06:45.590 --> 06:51.350 You can see it is progressing, the Delta T is progressing and it is not same as what we had defined 06:51.350 --> 06:58.130 because we have set the maximum current number and we also have the delta t, so it can alter the delta 06:58.130 --> 07:02.240 T to, uh, adapt to the maximum number. 07:02.240 --> 07:04.160 That's what adjustable time step is. 07:04.160 --> 07:06.200 It's mostly adaptive time stepping. 07:06.920 --> 07:10.640 Also you can monitor at what time the simulation is happening. 07:10.640 --> 07:14.150 Like this is 0.018 and so on. 07:14.150 --> 07:19.700 And we can see the parameters residual parameters for UX UI. 07:20.000 --> 07:25.550 And we also have for epsilon k pressure continuity everything. 07:25.550 --> 07:27.080 So this goes on. 07:27.770 --> 07:32.870 And finally when it reaches 0.3 this simulation will end. 07:32.960 --> 07:40.490 So when you are doing this uh you can actually go to that location and see it. 07:41.000 --> 07:43.700 So I will tell you what that latest time is all about. 07:43.700 --> 07:48.940 Now, if I want to stop this simulation, I can just do control C and it will just end there. 07:48.970 --> 07:51.970 Now I will do explorer dot x. 07:52.750 --> 07:59.020 So this location is open and you can see we have all the time steps coming here from 0.01. 07:59.020 --> 08:00.400 And it is slowly progressing. 08:00.400 --> 08:03.250 We have up to 0.14. 08:03.430 --> 08:10.900 So since we have given latest time now see the last time which it ran was 0.141795. 08:10.960 --> 08:15.880 So when I run the pimple form again it will just start from 0.14. 08:16.360 --> 08:20.470 See, it is starting from 0.14 because that is the latest available time step. 08:20.470 --> 08:23.950 So it is starting to write from 0.15. 08:23.950 --> 08:29.830 That is the advantage of having the start time as the latest time instead of starting. 08:30.700 --> 08:38.080 Okay, now the simulation is over as you can see, it took so much time to finish this though I have 08:38.080 --> 08:45.430 got all the, uh, running time in the video, so it took a lot of time and you can see how many actual 08:45.430 --> 08:47.590 seconds it took here. 08:47.590 --> 08:48.040 Okay. 08:48.310 --> 08:50.050 You can also monitor that. 08:50.320 --> 08:54.040 And you can see it ended at 0.3 okay. 08:54.040 --> 08:57.370 And maximum delta T was around 0.0002. 08:57.370 --> 09:03.220 So that was much greater than what we defined like 0.0001 was what we defined. 09:03.430 --> 09:07.510 And you can see a maximum current number is just five here. 09:07.510 --> 09:10.570 So that's why it got adopted okay. 09:10.570 --> 09:14.290 Now we have all the, uh files here. 09:14.290 --> 09:16.480 I'll open the paraview. 09:26.920 --> 09:28.870 Okay, now it has loaded. 09:28.900 --> 09:33.310 I'll click on apply and I'll change it to velocity. 09:33.550 --> 09:34.090 Okay. 09:34.090 --> 09:37.990 Now as you can see we are at 0.01 flow time. 09:38.140 --> 09:45.850 It is always advisable to first go to the last frame and click on this thing which is rescale to visible 09:45.850 --> 09:49.500 data range, and then go back to your first time step. 09:49.920 --> 09:57.480 Mostly that would be better if you are running a transient cases, but if you think the physics is like 09:57.480 --> 10:03.120 the initially, it will be high in velocity and finally it will be lower in velocity. 10:03.120 --> 10:07.350 Then you have to do the rescale to visible data range in the first time step. 10:07.380 --> 10:15.210 So in this case the final time step has some less velocity than the first time step. 10:15.210 --> 10:20.730 So first time step has to be rescaled so that it can adapt till the last time step. 10:20.820 --> 10:25.050 So you will understand when you are actually working on multiple cases and you will get it. 10:25.380 --> 10:27.540 So I will click on play. 10:28.410 --> 10:31.320 So this is how the flow is progressing. 10:31.440 --> 10:38.520 So you can see the time is changing here I'll click play again and it will start from 0.00 0.01 seconds. 10:41.100 --> 10:44.580 Now as you can see the minimum is 0.01. 10:44.580 --> 10:51.120 But if you want to include the zeroth time timestep as well, you can check on this button and click 10:51.120 --> 10:51.690 apply. 10:51.900 --> 10:57.300 Now you would also have the zeroth timestep, but mostly it wouldn't make sense at all because it is 10:57.300 --> 10:58.650 just initial condition. 10:58.800 --> 11:02.310 Click on run and it will be done. 11:02.310 --> 11:07.200 So it is not generally advised to take the zeroth time step because it is just initial values. 11:07.200 --> 11:10.080 It won't, uh, give you any proper insight. 11:11.640 --> 11:13.950 Okay, now we got the velocity. 11:13.950 --> 11:21.990 Similarly, you can also monitor pressure and go to the first time step, rescale it and click on pressure. 11:21.990 --> 11:25.440 And you can see it is getting like an animation. 11:26.310 --> 11:32.310 Now I can do this same thing for k nu t epsilon. 11:32.430 --> 11:34.440 Okay so you get the point. 11:34.530 --> 11:40.890 Now if I want to do uh velocity streamlines as an animation then I can still do it. 11:43.080 --> 11:48.350 I just plot streamlines, just like how, uh, just the way I would do normally. 11:48.680 --> 11:50.000 I'll hide the line. 11:50.390 --> 11:50.570 Okay. 11:50.570 --> 11:53.120 Maybe reduce the number of lines here. 11:55.910 --> 12:02.930 Okay, now I can do the same thing, like start play and it is running. 12:02.930 --> 12:11.030 Now it is a bit slow and loading because it is, uh, loading all the time steps and plotting the stream 12:11.030 --> 12:11.300 lines. 12:11.300 --> 12:13.580 So it is doing multiple things at the same time. 12:13.580 --> 12:15.830 So it is a bit glitchy. 12:15.920 --> 12:24.080 So what you can do is go to file, click on Save Animation and probably I will save it as stream lines. 12:25.220 --> 12:30.260 Hit okay and probably I will take yeah. 12:30.260 --> 12:33.560 Just for this case I would take just one frame per second. 12:34.190 --> 12:34.430 Okay. 12:34.730 --> 12:42.980 But if you want a realistic scale, then you have to choose ten frame per seconds for, uh, for you 12:42.980 --> 12:47.510 to get a slowed down animation of ten times. 12:47.510 --> 12:53.540 But if you do 100 frames per second, then you would get the exact realistic thing, because our lowest 12:53.540 --> 13:00.890 time step is 0.01 and our delta T is, uh, I mean delta t in terms of saving animation. 13:00.890 --> 13:04.670 Every frame here as we saved it is 0.01 seconds. 13:04.670 --> 13:08.900 So when you want everything for one second, you just have to multiply it by 100. 13:09.620 --> 13:13.310 But that would be too fast because the animation would end in 0.3 seconds. 13:13.310 --> 13:15.920 So I'll change it to one and click on okay. 13:17.000 --> 13:19.580 So you don't have to worry about this error. 13:19.580 --> 13:23.870 It is uh because of the resolution okay. 13:23.870 --> 13:24.710 It doesn't matter. 13:24.740 --> 13:25.220 now. 13:25.220 --> 13:32.630 We got the file, saved the animation file, and I can open the stream lines and it would be decent 13:32.630 --> 13:33.710 enough and fine. 13:33.710 --> 13:37.700 As you can see, it is saving one frame per one second. 13:39.170 --> 13:47.330 You can get some good insight from this, but if you really want, uh, a fast forwarded kind of thing, 13:47.330 --> 13:54.670 like a realistic animation kind of thing, maybe a ten times slow but animation kind of thing. 13:54.670 --> 13:59.830 Then you can choose ten frames per second in this case. 13:59.830 --> 14:01.870 Okay, I'll just save it. 14:12.460 --> 14:14.320 Now I will run this. 14:15.550 --> 14:18.910 Now you can see it is much better and looking like an animation. 14:19.390 --> 14:23.320 Since it is a streamline, we can't infer much as well. 14:23.830 --> 14:28.810 I will go here and save this as, uh, velocity file. 14:30.520 --> 14:37.570 But instead of ten I am even going to use 20, so it is just half scaled down. 14:39.910 --> 14:40.660 Go here. 14:40.660 --> 14:41.980 Run the velocity file. 14:44.110 --> 14:46.900 So this ran very fast as you can see. 14:47.920 --> 14:52.060 So this is what a real animation would look like at the end. 14:52.060 --> 14:57.640 If we had much higher time steps of flow time then we would have a longer animation. 14:57.640 --> 14:59.830 But this was just for demonstration. 14:59.830 --> 15:00.940 So this is okay. 15:02.050 --> 15:07.210 Now, you understood how to set up a transient case and run this. 15:07.210 --> 15:09.970 So this is the first part of the video. 15:10.420 --> 15:13.000 Uh, from the second part we would continue. 15:13.000 --> 15:13.450 Okay. 15:14.890 --> 15:15.250 Okay. 15:15.250 --> 15:17.650 Now we are going to see decomposing methods. 15:17.650 --> 15:27.820 So decomposing is a method where you will split the mesh into separate sections or regions, and give 15:27.820 --> 15:32.230 your processors each part of the mesh and ask it to solve for you. 15:32.230 --> 15:40.000 So this is done through a process called message passing interface, also popularly known as MPI. 15:40.240 --> 15:42.580 Openfoam comes with the MPI installation. 15:42.580 --> 15:48.340 Generally, people have to write MPI codes to parallel process their simulations. 15:48.340 --> 15:52.350 But since Openfoam comes with the installation of MPI, we can use it. 15:52.470 --> 15:55.530 So it also depends on the number. 15:55.530 --> 15:57.570 Of course, your system has. 15:57.570 --> 16:01.890 So cores are the number of physical processors your CPU comes with. 16:01.920 --> 16:02.280 Okay. 16:02.280 --> 16:08.310 So your CPU will have certain number of physical cores and logical course Openfoam can access only physical 16:08.310 --> 16:08.610 cores. 16:08.610 --> 16:15.330 Commercial software like Ansys fluent can access even logical processors, but Openfoam can access only 16:15.330 --> 16:16.500 the physical cores. 16:16.500 --> 16:22.020 So we will see how to find out the physical cores in your PC, because this is the number or value which 16:22.020 --> 16:24.600 you have to remember while decomposing. 16:25.200 --> 16:29.850 So I have given a set of methods to find out the course in your PC. 16:29.880 --> 16:35.820 So you just have to open your task manager and you will go to the performance section. 16:35.820 --> 16:38.460 And inside that you will go to CPU section. 16:38.490 --> 16:42.540 Under there you will go to course and you can find the number of course. 16:42.540 --> 16:44.430 So you just have to remember it. 16:44.910 --> 16:45.150 Okay. 16:45.150 --> 16:48.840 Sometimes your laptop comes with the stickers saying uh number. 16:48.840 --> 16:54.360 Of course if it is a gaming laptop, but if you are on a very regular laptop, then mostly it won't 16:54.360 --> 16:54.600 be there. 16:54.600 --> 16:55.800 So you have to find it out. 16:56.880 --> 17:04.590 So to do that, either you can open task manager from this or any shortcut methods which you know, 17:04.590 --> 17:06.510 there are a lot of shortcuts for this. 17:06.990 --> 17:10.350 So I'll go here, click on performance and on CPU. 17:10.380 --> 17:12.750 As you can see we have the number of cores here. 17:12.750 --> 17:14.400 In my system it is eight. 17:14.400 --> 17:21.930 So whenever I'm using eight for decomposing and MPI please remember that you have to use the num value 17:21.930 --> 17:24.360 which your PC is supposed to have. 17:24.510 --> 17:31.020 Just because I am typing eight somewhere, please don't follow the same and uh, face errors, so just 17:31.020 --> 17:33.240 replace it in the appropriate places. 17:33.630 --> 17:35.190 I will close the task manager now. 17:36.870 --> 17:39.660 Okay, now we are going to see the decomposing methods. 17:39.660 --> 17:46.650 So decomposing in open form is done through a file called decompose uh part dict okay. 17:47.160 --> 17:48.600 Uh we will see the file as well. 17:48.600 --> 17:50.700 So this is how the file actually looks like. 17:50.700 --> 17:52.490 It will have number of subdomains. 17:52.730 --> 17:55.040 And it will ask you for a method. 17:55.160 --> 18:00.320 If you choose hierarchical then you have to define the coefficients. 18:00.320 --> 18:08.210 So the coefficients are like if you have an entire mesh and in each direction how many subdivisions 18:08.210 --> 18:08.930 do you need. 18:09.650 --> 18:18.590 So in this example we are subdividing the x in four subdivisions the Y into two subdivisions, and the 18:18.650 --> 18:22.220 Z is going to be just one, which means it might be a 2D case. 18:22.220 --> 18:25.730 We are interested in decomposing only x and y. 18:25.760 --> 18:31.220 So what is important about the 4211 and why did I choose it? 18:31.220 --> 18:37.670 It is like when you multiply x by y by z you have to get the number of processors. 18:37.670 --> 18:43.880 So in this case it is four times two times one which is eight which is the same as number of subdomains. 18:44.150 --> 18:48.380 It is not mandatory that you always have to use all the cores which you have. 18:48.410 --> 18:56.750 If you have eight cores, you can still use just six or just two or just, uh, four. 18:56.750 --> 18:57.020 Okay. 18:57.020 --> 19:02.600 But it is very recommended that you use even numbers and not odd numbers. 19:03.110 --> 19:05.900 So use two, four, 6 or 8. 19:06.200 --> 19:06.980 Things like that. 19:06.980 --> 19:09.290 But don't use odd numbers. 19:09.590 --> 19:11.900 Mostly it can't work well. 19:11.900 --> 19:16.250 That is why laptops or computers come with even numbers, of course. 19:16.280 --> 19:18.680 Okay, so you always use even numbers. 19:18.680 --> 19:24.380 If you want to find out why it has to be even numbers, you have to study about message passing interface. 19:24.920 --> 19:25.250 Okay. 19:25.250 --> 19:30.740 Now in uh, other method which is called scotch, you don't have to define the coefficients. 19:30.740 --> 19:33.410 The program will do it for you. 19:33.410 --> 19:37.160 So you just have to specify the number of subdomains and the method. 19:37.970 --> 19:44.540 So to execute the decompose predict file and decompose your case the command is just decompose part 19:44.930 --> 19:45.290 okay. 19:45.290 --> 19:50.060 The purpose of decomposing is to divide the computational domain into smaller parts. 19:50.060 --> 19:55.960 The benefit is it enables parallel processing and reduces computational time by splitting the work to 19:55.960 --> 19:57.130 all the processors. 19:57.220 --> 20:04.090 When you are running in series, the computation will be done in a series manner by one processor, 20:04.090 --> 20:06.610 but you are doing parallel processing. 20:06.610 --> 20:11.650 The work will be split to multiple processor and it runs in parallel, so you get it faster. 20:12.610 --> 20:17.770 So the method is like first you split the domain into sub domains and assign different tasks to each 20:17.770 --> 20:22.120 processor, and you will be able to run the simulation or mesh. 20:22.120 --> 20:24.430 So the Openfoam command is decomposed part. 20:24.790 --> 20:30.160 Okay, so before we jump into what parallel processing and how to do it just we will see how to do the 20:30.160 --> 20:31.180 decomposing. 20:31.780 --> 20:35.020 I'll go to the terminal. 20:35.020 --> 20:39.880 I'll clear this and go to the second tutorial of today. 20:41.650 --> 20:48.460 Okay, we have zero constant and system as usual, but if you see under system we have one more file 20:48.460 --> 20:50.020 called decompose bardic. 20:50.020 --> 20:52.540 So that is what we are going to view now. 20:52.540 --> 20:56.680 Okay I will go to the second folder. 20:57.130 --> 20:58.840 We have decompose bardic. 21:02.890 --> 21:08.800 Here initially I have chosen Scotch, but if you want to trial with hierarchical, you can do it. 21:08.800 --> 21:12.190 So we have the number of subdomains where I have specified eight. 21:12.190 --> 21:13.570 You have to specify the number. 21:13.570 --> 21:14.560 Of course you are. 21:16.120 --> 21:18.790 And the method in this case is scotch. 21:18.790 --> 21:24.730 Otherwise it is just normal daily case which we run here for transient phenomena. 21:25.180 --> 21:27.220 We will see the control dict. 21:28.450 --> 21:36.040 So we are again using pimple form only and the time step is delta t is equal to 0.0001. 21:36.340 --> 21:38.980 And let the maximum see will be five. 21:38.980 --> 21:42.760 Sorry, this shouldn't be there. 21:42.760 --> 21:52.360 So that was my mistake okay so we have the delta T as 0.001 and maximum current number as five. 21:52.570 --> 21:53.020 Okay. 21:53.020 --> 21:54.910 So we are going to decompose it. 21:54.910 --> 21:55.300 Now. 21:55.300 --> 21:58.900 As you can see initially we have only zero constant and system. 21:58.900 --> 22:03.250 So this zero file has u p nu t k and epsilon. 22:04.120 --> 22:09.610 Because we are using k epsilon turbulence model, all you need as turbulence parameters is just these 22:09.610 --> 22:12.220 three turbulence viscosity k and epsilon. 22:12.460 --> 22:16.840 If you want to explore what I said for the boundary condition, you can go in and check it. 22:16.840 --> 22:18.990 But that's not the scope of this video. 22:19.080 --> 22:21.450 Okay, so we have zero. 22:21.840 --> 22:24.330 We don't have any other time step because we didn't run it. 22:24.390 --> 22:28.800 Now I will go here and I will execute decomposing. 22:28.800 --> 22:32.490 So the command is decompose part with P capital hit enter. 22:33.840 --> 22:37.290 As you can see it is saying that it is subdividing. 22:38.040 --> 22:38.250 Okay. 22:38.250 --> 22:42.270 It is subdividing the time zero into 0 to 7 which is eight. 22:42.750 --> 22:43.830 uh, field transfer. 22:43.830 --> 22:49.080 And you can see the number of cells, number of points has been assigned to each processor. 22:49.080 --> 22:51.420 So it might be a little bit more or less. 22:51.420 --> 22:59.490 But each processor has a set of mesh which it has to, uh, work on and do the simulation. 22:59.490 --> 23:00.000 Okay. 23:00.000 --> 23:06.090 So after we do this we got these files processor zero to processor seven. 23:06.240 --> 23:13.980 So when you go into processes zero you have made a copy of process uh zero files and the constant file 23:13.980 --> 23:20.340 under Polymesh you will have only the mesh this processor is assigned to you. 23:20.340 --> 23:23.430 It will not have all the mesh which you generated using block mesh. 23:23.430 --> 23:28.320 It will have only the mesh content which is necessary for this processor. 23:28.320 --> 23:30.870 And same applies to every processor Okay. 23:30.870 --> 23:35.970 So it has the zero file because we are going to run things in parallel and it needs a reference for 23:35.970 --> 23:36.420 zero. 23:36.420 --> 23:38.490 So it has the zeroth time step. 23:38.670 --> 23:43.260 Now I will explain what parallel processing is and how to do it. 23:44.760 --> 23:47.220 So parallel processing using MPI. 23:47.280 --> 23:50.040 So MPI is a message passing interface. 23:50.160 --> 23:53.280 The definition is it is standard for parallel computing. 23:53.280 --> 23:58.170 And the functionality is functionalities it allows processes to communicate with each other. 23:58.470 --> 24:04.470 In implementation for Openfoam, we will enable parallel processing for faster simulation, which is 24:04.470 --> 24:06.810 the same for any CFD software. 24:07.050 --> 24:11.280 And you will, uh, run a piece of foam simulation for this. 24:11.910 --> 24:19.890 Uh, as an example for yourself, you will try a piece of foam tutorial if you want and you can try 24:19.890 --> 24:21.000 it using MPI. 24:21.030 --> 24:25.470 For now, we will be sticking with the simple form tutorial and we will see a demonstration. 24:25.470 --> 24:31.770 But if you want to do it with icofom or simple form, the methodology remains the same. 24:31.770 --> 24:40.020 All you have to do is you have to copy this decomposed par dict and put it into any system directory 24:40.020 --> 24:42.270 of any case, and you can decompose it. 24:42.270 --> 24:46.220 You just have to set everything properly like the number of subdomains. 24:46.250 --> 24:46.640 Okay. 24:46.640 --> 24:49.220 So that is the only thing which you have to do. 24:50.870 --> 24:51.470 Okay. 24:51.470 --> 24:55.130 The benefit is it utilizes multiple processes simultaneously. 24:55.130 --> 24:58.610 And it scales well with increasing computational resources. 24:59.780 --> 25:04.580 So let's see what is the step to do parallel processing using MPI. 25:04.820 --> 25:07.250 The first thing is decompose power. 25:07.250 --> 25:11.820 But but before that you must have a mesh and initial boundary condition in zero five. 25:12.240 --> 25:16.350 So that's what I mentioned as poly mesh must be available because not just block mesh. 25:16.350 --> 25:21.420 You could have generated that with snappy hex mesh or imported from Ansys fluent mesh, which we will 25:21.420 --> 25:21.840 see later. 25:21.840 --> 25:25.470 So anyways, it will all the mesh will be in polymer, so it has to be there. 25:26.070 --> 25:32.060 Then you are running decompose power which will subdivide everything, every mesh for each process. 25:32.600 --> 25:40.220 Then the command to initiate the solution or simulation is MPI run space hyphen NP space. 25:40.220 --> 25:46.760 The number of processor, which in my case is eight and uh process which is in this case pimple foam. 25:46.760 --> 25:49.370 If it is a simple form then you have to do simple form. 25:49.370 --> 25:55.610 If it is a micro form you have to type ICO form in place of process and space hyphen parallel. 25:55.640 --> 25:59.780 So this is what will initiate the solver to run in parallel. 26:00.560 --> 26:01.730 I have put some examples. 26:01.730 --> 26:08.540 If it is an eight core processor, it will be mpirun space hyphen NP space eight and the command if 26:08.540 --> 26:09.290 it is simple form. 26:09.440 --> 26:11.240 Simple form space hyphen parallel. 26:11.990 --> 26:15.980 If it is six cores, then NP six. 26:16.040 --> 26:17.210 Simple form parallel. 26:17.480 --> 26:22.390 Okay, we will see about reconstruct, uh, after we run the simulation, actually. 26:22.690 --> 26:22.930 Okay. 26:22.930 --> 26:31.270 Now that we have all the decomposed files, we can initiate it through Mpirun NP eight. 26:31.300 --> 26:31.630 Eight. 26:31.630 --> 26:35.200 In my case, simple form. 26:37.360 --> 26:37.930 Parallel. 26:38.860 --> 26:45.100 When I hit enter, it will start to do the same process, which we did previously, but in parallel. 26:45.100 --> 26:47.140 So it is going to end super fast. 26:48.730 --> 26:50.590 As you can see, it is running super fast. 26:50.590 --> 26:54.490 Then what it was running previously because it is just eight times faster right now. 26:59.770 --> 27:02.830 So let's wait for the simulation to end and I will get back to you. 27:06.520 --> 27:09.010 So the simulation is about to end. 27:09.040 --> 27:10.840 Okay, it is done. 27:11.200 --> 27:14.500 Now I will do ls again. 27:14.500 --> 27:16.540 All you can see is processes 0 to 7. 27:16.540 --> 27:18.190 We are not able to see the time step. 27:18.640 --> 27:21.820 But what happened because everything is inside processor. 27:22.150 --> 27:28.870 So we have 0.1 to 0.3 everything here for every processor. 27:29.680 --> 27:30.100 Okay. 27:30.100 --> 27:36.730 So we have to reconstruct everything and then we can, uh, see the visualization. 27:37.060 --> 27:42.640 So what reconstructing will do is it will take the mesh information from every processor, collectively 27:42.640 --> 27:46.930 put it, and just save everything as, uh, separate time step. 27:46.930 --> 27:49.180 And then it would look something like this. 27:49.180 --> 27:57.310 But if you don't want to reconstruct because it takes a bit of time, you can open paraview Paraview 27:58.060 --> 28:01.510 and instead of directly clicking on apply, you can change it here. 28:01.510 --> 28:06.760 Instead of reconstructed case, you can go to decomposed Case and then click on apply. 28:07.930 --> 28:08.350 Okay. 28:08.350 --> 28:11.950 Now what if I just click on reconstructed case. 28:12.100 --> 28:17.200 You are not going to see anything because it doesn't know that if there are any files at all, because 28:17.200 --> 28:21.040 it is expecting the time 0.01, but it is not there. 28:21.040 --> 28:23.250 So you have to give decomposed case. 28:23.250 --> 28:27.150 So it will go inside every processor and try to recollect it for you. 28:27.420 --> 28:28.380 For visualization. 28:28.380 --> 28:30.240 Now I can see the velocity. 28:31.320 --> 28:36.090 Right now if I play, it is going to be playing because it is trying to recollect. 28:36.090 --> 28:39.930 But since the mesh is very less, mesh count is less. 28:39.930 --> 28:41.970 We are not getting stuck. 28:41.970 --> 28:47.370 But when the mesh count is very high, let's say in a 3D case where where there are millions of cells, 28:47.370 --> 28:50.670 then every loading time step is going to be very high. 28:50.700 --> 28:55.290 That's why we are going to reconstruct and then visualize it very easily. 28:55.470 --> 28:55.710 Okay. 28:55.710 --> 29:01.560 Because it takes time for paraview to reconstruct every single time step without storing it and displaying 29:01.560 --> 29:02.100 it to you. 29:02.100 --> 29:06.570 It's too much on the memory, so we will see how to reconstruct it. 29:06.750 --> 29:11.520 To reconstruct the very simple thing is just run, reconstruct part. 29:11.640 --> 29:13.230 That's all you have to do. 29:13.230 --> 29:16.860 So just like how you did decompose part, now you have to do reconstruct part. 29:17.310 --> 29:17.460 Okay. 29:17.460 --> 29:21.870 Now I will go to my terminal and I will do reconstruct part. 29:21.870 --> 29:23.340 And you will see what happens. 29:23.340 --> 29:29.910 It is starting to reconstruct every single time step by putting the time step name. 29:29.940 --> 29:34.350 See it is saying time 0.01 and it is reconstructing all the fields. 29:34.470 --> 29:37.290 Volume, scalar field, vector field and surface scalar fields. 29:37.470 --> 29:40.170 So as you can see it is taking a bit of time. 29:40.170 --> 29:41.280 It is running slow. 29:42.330 --> 29:46.170 We will also see a method to run this parallel as well. 29:46.530 --> 29:48.990 But for now this is the most popular method. 29:48.990 --> 29:51.150 So let it run. 29:51.150 --> 29:56.220 We are not going to wait for it to end completely because this is just for demonstration. 29:56.220 --> 29:58.140 We will stop it at 0.05. 29:58.170 --> 29:59.190 Till then, let's wait. 30:08.550 --> 30:13.500 Okay, now it is done before it starts with 0.06. 30:13.500 --> 30:20.970 If you want to end any process, just hit Ctrl C and it will be interrupted by the keyboard. 30:21.510 --> 30:26.390 If I go here now, you can see we have 0.01 to 0.05. 30:27.440 --> 30:32.150 Okay, but let's assume that you have reconstructed everything, okay? 30:32.150 --> 30:41.090 And, uh, you don't need these files anymore because these are going to consume a lot of space. 30:41.090 --> 30:43.040 Since this is a very small scale, it is less. 30:43.040 --> 30:44.930 Otherwise it's going to be in gigabytes. 30:45.350 --> 30:49.250 So we don't want these anymore because the data has been reconstructed. 30:49.250 --> 30:54.620 So what you can do is arm space minus r proc star. 30:54.620 --> 31:00.560 So it just says that delete everything and everything inside those folder which starts with the four 31:00.560 --> 31:01.940 letters p, r o, c. 31:01.940 --> 31:04.010 And we don't care what comes later. 31:04.010 --> 31:07.700 So anything any folder that starts with proc will be deleted. 31:07.820 --> 31:13.520 When you're doing through terminal, it's going to be faster than what you will be manually clicking 31:13.520 --> 31:15.020 and deleting through windows. 31:15.020 --> 31:22.220 Since this is through WSL, it's going to be faster than what windows can even do. 31:22.370 --> 31:25.340 So as you can see, it is slowly getting deleted one by one. 31:29.330 --> 31:31.040 And now it is all done. 31:31.040 --> 31:36.470 Okay, now you can go to paraview and click on Reconstructed Case. 31:36.470 --> 31:38.390 Click on apply and it will still work. 31:38.390 --> 31:43.460 But we have only 0.05 seconds of data. 31:44.120 --> 31:48.230 Okay, now let's say like you don't want these time step. 31:48.230 --> 31:55.850 Also, you want to start from uh scratch by changing let's say delta t or maximum quota number if you 31:55.850 --> 31:57.500 want to choose it as two. 31:57.500 --> 32:00.230 And if you want to run it, you have to delete all the time steps. 32:00.230 --> 32:04.910 Right now it is just like, uh, around 100 time step. 32:04.910 --> 32:09.560 But what if it is so much you can't delete everything one by one? 32:09.560 --> 32:14.600 Or just even if you select everything and delete windows is going to take a lot of time. 32:14.720 --> 32:22.160 So Openfoam gives you this nice command for list times. 32:22.460 --> 32:26.650 Notice that L and T are capital space hyphen m. 32:26.650 --> 32:31.480 Hit enter and all the time steps will be gone, like in an instant. 32:31.660 --> 32:34.390 So you have deleted all the time steps through this command. 32:34.390 --> 32:38.920 So this is just some additional information apart from today's topic. 32:39.940 --> 32:45.190 So this is just nice to know because when we are working with terminal, it will be much faster than 32:45.190 --> 32:46.870 what you can do directly with windows. 32:48.070 --> 32:48.220 Okay. 32:48.220 --> 32:54.430 now we know how to decompose a case and run it in parallel using MPI and also reconstruct part. 32:56.020 --> 33:00.790 Okay, now we are going to see how to run parallel snappy hex mesh. 33:01.030 --> 33:03.130 Snap hex mesh is a meshing process. 33:03.130 --> 33:09.160 And we are going to do meshing itself as uh parallel processing. 33:09.160 --> 33:15.760 As I was saying, we must have a poly mesh to generate any sort of decomposing, right? 33:15.760 --> 33:20.680 Because we need a poly mesh to even put it in constant and split the mesh to the processor. 33:20.980 --> 33:23.620 But snappy mesh itself is a mesh. 33:23.620 --> 33:29.140 So what we are going to do is we will generate the block mesh and split the block mesh to different 33:29.140 --> 33:34.990 processor, and we will do snappy mesh as a process to every processor. 33:35.230 --> 33:40.840 So I think you get the point and we will do the tutorial so you understand it better. 33:40.870 --> 33:42.100 I'll clear this. 33:42.820 --> 33:46.390 Go to the third case for this class. 33:47.050 --> 33:47.290 Okay. 33:47.290 --> 33:52.300 Now we just have constant and system no zero file because we are not interested in running simulation. 33:52.750 --> 34:00.280 So just this now what I'm going to do is I'm going to do block mesh because first we are going to run 34:00.280 --> 34:01.030 the block mesh. 34:01.030 --> 34:07.960 So this is the pipe external case which we made if you remember meshing around the pipe. 34:08.080 --> 34:08.440 Okay. 34:08.440 --> 34:10.240 Now I have generated the block mesh. 34:10.360 --> 34:13.780 Then the next command is to decompose this. 34:14.650 --> 34:20.500 So I will decompose using decompose part. 34:20.500 --> 34:20.500 Ah. 34:28.930 --> 34:31.360 Okay, now we have decomposed the block mesh. 34:31.390 --> 34:34.990 Now, the next thing is we have to initiate snap x mesh. 34:35.710 --> 34:40.630 So the command is same Mpirun NP eight. 34:41.560 --> 34:45.220 And the process name which is snappy x mesh. 34:45.400 --> 34:46.810 As usual parallel. 34:46.810 --> 34:53.970 If you remember we always do snappy x mesh with space overwrite, so we can also include that it doesn't 34:53.970 --> 34:57.600 matter if you put parallel first or overwrite first, it's going to be the same. 34:57.600 --> 35:01.860 So what I have done is let me shorten this so that you can see it better. 35:05.820 --> 35:06.330 Right. 35:09.480 --> 35:09.960 Okay. 35:09.960 --> 35:15.720 It is mpirun I am not able to put it in the a single line or can I? 35:17.580 --> 35:18.090 Yeah. 35:18.210 --> 35:23.280 So it is just like MPI run PNP, snappy hex mesh, parallel override. 35:23.640 --> 35:28.740 Now I can hit enter and this will start the snappy hex mesh process. 35:29.730 --> 35:31.710 Now you see this running much faster. 35:31.710 --> 35:37.020 If you remember at what speed it was running when you initiated the mesh, you would know that it is 35:37.020 --> 35:38.460 much, much faster. 35:39.390 --> 35:47.220 So now let's wait for it to end and we will reconstruct the mesh through a special command called reconstruct 35:47.220 --> 35:47.970 per mesh. 35:48.000 --> 35:55.140 So as you remember, we just did reconstruct part when we are doing the time step reconstruction. 35:55.140 --> 36:00.360 But in this case we have to do reconstruct per mesh space hyphen constant. 36:00.360 --> 36:04.260 Because we are trying to write this mesh inside the constant folder. 36:04.380 --> 36:12.630 So I will do reconstruct power mesh hyphen constant. 36:13.140 --> 36:13.920 I'll hit enter. 36:13.920 --> 36:18.750 Now you can see it is taking data from all eight processors and putting it in the. 36:20.790 --> 36:21.420 Poly mesh. 36:21.450 --> 36:26.250 Okay now we don't need these processor 0 to 7 anymore. 36:26.250 --> 36:30.480 So what we can do is r minus r proc. 36:32.040 --> 36:34.530 And it will delete all the processor files. 36:34.530 --> 36:36.180 So we can save upon some space. 36:36.180 --> 36:41.610 Now if you just want to view the mesh you can go to para Para view. 36:42.390 --> 36:43.470 Click on apply. 36:45.600 --> 36:48.480 You can go to wireframe and you can see we have the mesh. 36:48.720 --> 36:52.910 So the output is going to be the same but the process will be faster. 36:52.910 --> 36:55.310 That is the advantage of using MPI. 36:56.450 --> 37:05.120 So I hope you understood how to do the uh decomposing as well as reconstruction. 37:05.120 --> 37:09.200 Now let's say you have a zero file also. 37:09.200 --> 37:11.630 And now you want to run things in parallel. 37:11.630 --> 37:17.300 So what you can do is after deleting all the processor files now, then reconstruct, just like how 37:17.300 --> 37:21.080 you would do for any simulation case like this process. 37:21.080 --> 37:22.700 So again you have to decompose. 37:22.700 --> 37:25.700 Then you can run things in parallel and reconstruct part. 37:25.880 --> 37:27.710 So that is the best method. 37:28.460 --> 37:30.530 So now we have an additional tip. 37:30.530 --> 37:33.050 What if you want to reconstruct in parallel. 37:33.050 --> 37:36.950 Because reconstruct if it has like too many time steps. 37:36.950 --> 37:40.880 Like the previous case it had a lot of time steps, right? 37:40.880 --> 37:47.000 So if the case is like that, then it's going to take a lot of time to reconstruct. 37:47.000 --> 37:51.020 You saw how much time it took to reconstruct that. 37:51.020 --> 37:57.770 So what I will do is I will again run that case and show you how much time it takes. 37:59.180 --> 38:04.580 When we are reconstructing it in parallel, it's going to take a lot less time. 38:05.360 --> 38:09.830 So I'll just run this case and get back to you once this is over. 38:11.840 --> 38:14.090 I'm just going to run a few time steps. 38:14.090 --> 38:15.170 Not completely. 38:18.080 --> 38:24.440 So we have five time steps, or maybe four because I stopped it early. 38:25.640 --> 38:27.350 Yeah, we have five time steps. 38:27.380 --> 38:27.830 now. 38:27.830 --> 38:31.640 You remember how much time it took when we tried to reconstruct it? 38:31.640 --> 38:33.530 It was very, very slow. 38:34.130 --> 38:39.560 But now we are going to try to use this command and see how fast it is going to reconstruct, because 38:39.560 --> 38:40.430 it's running in parallel. 38:40.430 --> 38:43.940 So it is Mpirun NP eight again, the same thing. 38:43.940 --> 38:52.540 Then we are going to run a reduced tribute pa PA. 38:53.920 --> 38:56.620 Reconstruct. 38:58.030 --> 38:58.780 Parallel. 39:00.190 --> 39:02.590 New times. 39:02.680 --> 39:04.360 So you just type this. 39:04.360 --> 39:07.750 I hope I didn't make any mistake in spelling and hit enter. 39:07.750 --> 39:11.050 Now you see just how fast it is reconstructing. 39:19.450 --> 39:20.500 Now it is just done. 39:20.530 --> 39:23.950 This was the time it took to just reconstruct two time steps. 39:24.010 --> 39:29.050 So, uh, this is not very popularly known among a lot of users. 39:29.050 --> 39:32.110 So if you use this, it's an advantage to you. 39:32.650 --> 39:32.950 Okay. 39:32.950 --> 39:38.500 If you have any questions about anything related to this video, please feel free to contact me. 39:38.860 --> 39:40.150 See you in the next class. 39:40.150 --> 39:40.960 Thank you.