Annual Daylight Recipe - Large Model Feedback

Hello!

Last month I used Pollination to run a pretty large annual daylight simulation and would like to share the outcomes with everyone, along with some questions/suggestions:

Overview - The model was based on 4 residential towers with 20 stories each. I modeled the first two floors in detail, and the rest of the facade with generic surfaces. Since the towers are so closes to each other, I modeled all the facede windows to account for the impact on reflectance. The final honeybee model was made up of 9955 surfaces and 1556 apertures. I used 328 sensor grids, with a total of 21376 testpoints. Anyone interested in checking out the model in more detail, here is the URL: Pollination Cloud App . Here is a perspective of the model:

Results: The results for the simulation were pretty consistent with the same simulation I ran on my computer using ladybug tools, as expected. Since I was interested in looking at sDA 200 and not sDA 300, I had a bit of a hard time importing the results back into grasshopper and using the “HB Annual Daylight Metrics” componente. When downloading the results assets from pollination, I got 328 .ill files and one sun-up-hours file, which I had to combine in a tree structure. Not really that big of a problem, but it would be nice to able to get back theses result files from pollination in the same tree structure as we get using Lady Bug Tools to simulate locally.

Simulation Time: When simulating the same model on my Notebook (i7-10750H CPU @ 2.60GHz 2.59 GHz with 16Gb of RAM) , it took around 10 hours, which is ok considering I had a pretty time consuming Radiance Parameter setup. On Pollination however, it took more than 24 horas to run the same thing. I’m not sure how long it actually took, because on the pollination dashboard it says 17 minutes (I think it reset after 24 hours simulating, which would be a nice fix for larger models.) I didn’t tweek the sensor-count because I still don’t understand how to ajust it in a way that it really helps. Any suggestions?

Pollination Cloud APP: Everything worked pretty well on the cloud APP, with the exception of the simulation time bug I mentioned, and the online model visualization, which didn’t work with a model this big (the site just crashes).

Final Thoughts: Despite the problems I mentioned above, I really liked the new workflow and will use it for sure in future projects. I am looking forward to seeing the colaboration features and final projects\simulations folder structure, as it is one of the major headaches I have managing a 8 person simulation team in our firm. It would be nice to have a easier input-output relation between the pollination downloaded assets and the Ladybug Tools results componentes, especially for things that aren’t that comum, such as a sDA 200 analysis. Also, I know this has been mentioned in the Ladybug forum in the past, but it would be great if there was a out-of-the-box ASE analysis for LEED compliance!!

4 Likes

Hi @viniolmos, thank you for documenting this - very helpful.

The annual daylight recipe has an input for changing the threshold that should have saved you from downloading all the results.

@antoine can help you with this item. 24 hours doesn’t look right. From the start and end time it seems to be on the same day. Once we know more about this I can help with the sensor-count question.

We are aware of this issue and we already started working on an alternative solution. @devang, can probably provide you a HTML file with your model to test it.

Do you have an organization on Pollination? What parts of the collaboration do you need to know more about? @janki.vyas this is a topic that we should probably cover in our next user meeting and write a blog post that clarifies how Pollination makes collaboration effective.

I appreciate some clarification on final projects\simulations folder structure. How can we help with that?

Noted! We are working on this. The Option II will be available first but we are all set to also provide sDA and ASE with dynamic blinds. As I said it’s on our list.

Hi @viniolmos :wave:

Thanks for trying out Pollination and sharing back some feedback. It’s really exciting and helpful to see people trust the platform with real life use cases at this stage :raised_hands: As @mostapha mentioned I am probably the right person to help you enhance/figure out the performance issues you encountered.

We encountered some scaling issues and picked up on a couple of bugs during the month of June. We opened up Early Access specifically to better understand how our backend scales and how to deliver an optimal performance for end users. As such I think your Job was executed while we encountered some scaling issues…

I am 90% sure that the Job actually took 15 mins to complete. The nature of the bug we encountered between June 7th and June 20th meant that the status of Jobs was not updated in a timely manner for end users.

It would be great if you could re-run the same job again from your test-residential project page at some point and let us know how quickly it executes this time around. You can do this by clicking on the “New Job” button on your project page, selecting the recipe you used and then filling in the inputs arguments and files needed to execute your Job. The two images below show how to navigate the web app

Thanks again for testing out the platform and reporting back to us through Discourse. This feedback in super helpful!! :tada:

Hello Mostapha, thanks for the all the clarifications.

Since the early-access pollination cloud app only allows public projects, I’m interested in learning more about how this will work with private projects and various collaborators (answering your question, I am the only one in our firm trying out Pollination). One of our main challenges being a simulation team is the result and result file management, especially in projects with more than one modeler and different analysis types (daylighting and thermal comfort, which we generally separate in two models). Today we generally try to maintain a standard folder structure on our Google Drive for the results files and concentrate the different results on a spreadsheet, but it’s nearly impossible to do this effectively in a 8 person team. I was hoping that Pollination could help us out in that sense, at least with keeping track of the dates and users behind the result files. If you want to explore this topic more, we would happily schedule a call to explain in more details how we manage our models and results using ladybug tools.

Thanks!

1 Like

Hello @antoine !

I actually tried to do this and encoutered the same problem mentioned here when trying to select my model:

The model would not load in the Pollination Web App, so I had to start a new run opening the Grasshopper file. It is currently running, so I’ll keep you posted!

Thanks!

The simulation ran and the dashboard indicated a similar time. The last I checked, it was running for 15 hours!

Hi @viniolmos,

Sorry for the delayed response.

As @mostapha mentioned, we are already working on alternative solutions to visualize hbjson in the browser. We understand that visualizing results through grasshopper can be time-consuming for such large models. Therefore, these solutions will also support visualizing results in the browser. Following is what we’re working towards;

  1. Images at the end of simulations. We are working on providing the capability to generate images that will show the hbjson with results. Images below show the famous model that ships with Ladybug tools.



    You can imagine that for a large model such images will look too busy. Therefore, we are building a new viewer.

  2. A 3D viewer to visualize hbjson and results in a browser. Please download and double click on the html file that you will find in the attached
    viniolmos.zip (2.3 MB). In this html you will find your model with results. You can turn off or reduce the opacity of other model objects such as “walls”, “floor”, “ceiling”. Lastly, you will find “grid” object. Use the “color-by” dropdown to select the result. You can export images as well. I exported the following images from the viewer;




We are working hard so that users will be able to do all this and more without opening rhino + grasshopper. It goes without saying that feedback from users like yourself is invaluable. We do test our work at our end but there is nothing better than a user taking our work through real-world complexity. Thanks again for sharing your feedback with us.

2 Likes

Thanks @viniolmos for trying again!

@antoinedao and @tyler, can you double check and see if this job is really taking 15 hours or 15 minutes? If it is 15 hours then is this a bug in the frontend? If it is 15 minutes why would @viniolmos see it working for 15 hours? Something strange is happening here that we need to fix! :grinning:

1 Like

Hi @viniolmos et al!

I just checked the DB records and it indicates this job did take ~16min to finish. As @antoinedao mentioned, the first time this project was executed was before we had overhauled our message passing system to better handle large parallel executions. Are the results after 15 min not what you are expecting?

When you say:

Where are you checking that reports it is running for 15hrs?

It’s a bit difficult to say what’s going on without watching it happen, so I can set up a test of this project to debug to be sure we fix any issues!

2 Likes

I can confirm that the run itself takes only 14 minutes:

Here is how the model look like:

1 Like

Hi @mostapha and @tyler !

I tried a third run earlier today, and kept an eye on the Cloud APP and Grasshopper componente. It indeed took only 14 mn (which is really fast for such a large model), the results seem to make sense! Strange, because I did the same thing in the second run, and the grasshopper componente and the Cloud APP where showing ~15 hours when I stopped watching. I am going to try a few more runs and I’ll post some feedback later!

Thanks

3 Likes

Thank you @devang !

The 3D viewer looks great, look forward to seeing it being implemented!

2 Likes

Glad to hear it’s working as expected! And that it’s fast enough to make a difference :wink: .

We are making updates to the product constantly behind the scenes so it’s very possible that the specific run you were watching in GH before was not being updated properly if it was before the relevant fixes were live.

Let us know how it works out! (And the 3D viewer is coming soon…)

2 posts were split to a new topic: Recipe for LEED option 1 (sDA/ASE)