Hardware benchmarking and swap allocation

[Gabron]

Hi there,

I would like to share some hardware benchmarking for ODM processing. We used VMs (Google Cloud Engine) with a small survey (0.3ha with GSD 0.96cm) and our baseline ODM setup. Here a screenshot of the findings:

I have following questions:

1. 16 CPU cores with 64 RAM seemed to be fine and no GPU was needed; processing a survey of ~5ha would change these settings significantly?
2. What general hardware settings are being recommended by the ODM power users for processing up to 5ha surveys with 1cm GSD?
3. What would be the best swap allocation to a swap file for a 16/64 setting? According to a source, in most cases, when running Linux on a virtual machine, a swap partition is not present, so the only option is to create a swap file. Source: https://linuxize.com/post/create-a-linux-swap-file/

[smathermather]

@Gabron

16 CPU cores with 64 RAM seemed to be fine and no GPU was needed; processing a survey of ~5ha would change these settings significantly?

How many images are you estimating for your 5ha, 1cm survey?

[Gabron]

Hi Smarthermather, let me check back and get back to you tomorrow. Thanks for help on this!

[Gabron]

Hi smarthermather,

sorry for the delay of response. I found a survey for benchmarking. Here the specs:

• A survey for a field of 5.56ha had 3856 input images with an average GSD of 0.96cm. The area covered by the raw images is bigger than 5.56ha. Let me get the right metrics… Around 7ha are covered by these 3856 images.

• The survey was subdivided into 9 chunks

• We process in average 50K images per week (RGB + MS)

We want to switch from Pix4D to ODM. What hardware recommendation do you suggest, that has a good cost/benefit ratio?

Thanks for your help on this! Especially interested in your feedback, how inportant it is to have swap allocated and how much it should be. To allocate swap ressources was emphasized by @Saijn.

Best,
Gabor