Kessler@home is now in beta testing, and the first work has gone out.

The beta is 1,000 jobs: 100 scenarios, each run with 10 different random seeds. Every job goes to two volunteers' computers, and the two results have to agree exactly before they are accepted, so that's 2,000 tasks in all.

There are apps for CPUs on Linux, Windows and macOS, and for NVIDIA and AMD GPUs on Linux and Windows. Each computer can have up to 4 tasks per CPU thread and 10 per GPU at a time.

During the beta we're checking that results agree across all of these platforms, and watching run times, credit and errors. Work is limited to these beta tasks for now.

If a task fails or something looks wrong, please post in the message boards. Thank you for helping test Kessler@home!

Beta test #2 is underway to check the new app.

Congratulations on development of this project and contributions to Einstein&home! I look forward to detail of the science.
Do you know what the current plan is for total number of work units and simulations for the, I assume first phase in a long adventure?

---

weird, i thought only mods and admins could post here LOL. maybe i accidentally opened a setting.

anyway, the current plan is that the first phase will be about 120 million WUs, at least 240 million tasks. there's a LOT of work to do.

That is a lot; Twice the tasks as WU?
Will GPU be vastly more efficient than CPU? Would AVX optimized executable be beneficial?
Multiple WU per "wide" GPU technique be beneficial here as it is on Einstein?

2x tasks per WU is normal. in BOINC terms a WU is the collection of two tasks that get sent to two computers, and then the results of those two are compared for validation.

The GPU app is faster and probably more efficient, at least for the strong FP64 models, but I have made apps for most major platforms so you can contribute with whatever you have, big or small :) The CPU app is still very competent as the GPU app is not yet heavily optmized. I have only done a little optimization to make it useable. the app needs FP64 support.

yes multiple tasks per GPU will likely be helpful since one task is not ususally large enough to fully saturate the GPU, especially for GPU models like V100 that have a lot of FP64 power.

Understood, Quorum of 2, 240M ideally if no invalid results.
Over at LTT, TheLANguy stated about Kessler: "The project is closed source due to agreements with ESA/NASA". I read you work in the industry.
https://linustechtips.com/blogs/entry/2476-november-boinc-blog/

Actor/nerd David Hewlett has been interviewing on socials, ESA peeps for a while now, wink open the door for BOINC wink.

This is an independent project, and the source code for the app/sim model is closed at this time. However, I have posted a page that goes in-depth to exactly what validation I have done on the model before release. you can see that here: Project > Model Validation

The main reason for keeping it closed for now is an abundance of caution since the intent is to use the data analysis for a few research and PhD papers. full raw results data will probably not be available before the papers are written. I have some rough plans to form result summaries that can be posted to the website in a way similar to how I posted the Beta test results. I don't think anything in the simulation code itself prevents releasing it, but it's "mine" and I put a lot of work into fine tuning it and getting it to this point. There are certain things I truly cannot release though. mostly some raw/untouched data sources that were used to build/supplement the population data (ESA DISCOS), the population data sent with the apps are de-identified and derived outputs, which is allowed. and the ESA and NASA debris models themselves that were used for validation testing also have rules about redistribution, but I think they are mostly available if you follow the process create accounts with them and fill out the applications and sign the user agreements to get access to those models.

I have contacts with other partners and contractors in the CA/SSA industry, and would like to have one of them do a kind of 3rd party validation if I can. but that will be up to their time and ability to do so so I don't want to name names on that yet. It's not something that would prevent starting the sweep.

I understand some folk's hesitation with supporting a closed-source project, but I hope that I have shown enough validation testing data that people can trust me. AI tools also can and will be used by some and can pretty easily snoop the binaries to confirm what it actually does and is doing what I claim. Also consider that more BOINC projects are closed-source than open, so this is not really anything new, in my opinion.

Thank you for the support and consideration either way!

Ian