What happens with dumped WUs?

[foldinghomealone]

Hi all,

As I have stopped folding and just recently started over...
What happens when I intentionally "dump" a WU?
As I understand, a dump report is uploaded and no points are credited.

Sometimes I need my computer for other tasks and as long as the WU is small enough, I can wait till it's finished.
However, some CPU WUs take "forever" and I can't wait another few hours for letting it finish.

Will the partially calculated WU be uploaded and then finished by another user?
If not, why not?

[muziqaz]

They are dumped. No work is sent to the researcher. So basically, work is wasted.
If you feel that you cannot finish work units, consistently and cannot prevent that, please do not fold.
Thanks

[foldinghomealone]

I understand the reason behind requiring fully completed WUs before upload - e.g. to reduce network load. And I fully accept that partial uploads shouldn't receive any credit.

However, discarding valid partial results is inefficient and wastful - both in terms of energy and computing time.

Because the computing time of WU can vary significantly - and I have no control over which one gets assigned next - I often have to stop folding hours earlier than necessary just to avoid receiving a WU that might not finish in time.

If partial results could be uploaded - even without receiving any credit - I could continue folding right up until the last minute before I need my computer, instead of leaving the computer idle for playing it safe.

[muziqaz]

I understand the reason behind requiring fully completed WUs before upload - e.g. to reduce network load. And I fully accept that partial uploads shouldn't receive any credit.

However, discarding valid partial results is inefficient and wastful - both in terms of energy and computing time.

Because the computing time of WU can vary significantly - and I have no control over which one gets assigned next - I often have to stop folding hours earlier than necessary just to avoid receiving a WU that might not finish in time.

If partial results could be uploaded - even without receiving any credit - I could continue folding right up until the last minute before I need my computer, instead of leaving the computer idle for playing it safe.

FAH project expects a little bit of the responsibility from donors. We cannot control everyone's contributed time or hardware quality. We also cannot start splitting WUs in various sizes. That will be bigger waste to the resources than dumping incomplete WU. Each WU usually generates the next one server side. So if that WU is 95% complete and uploaded, as you suggest, and next person gets that remaining 5%, and that person's PC decides to pack up before finishing those last 5%. Now server still thinks work is being done and is waiting. So next WU is being delayed because of that last 5%. And then you have the issue of deadline and timeout for that remaining 5%, that would need to be calculated. Things get complicated fast. And don't get me started on the wasting time downloading and uploading that 5%, then joing to remaining 95%. That's server time, internet bandwidth, etc etc. Researchers have many many other things to do rather than putting jigsaw puzzles of their projects from random sized WUs. We like even numbers and full WUs
So, again, if you believe you cannot consistently contribute with your hardware, please don't. FAH will be fine. The fact that you tried to contribute already is greatly appreciated.

[foldinghomealone]

I understand what your're saying.
I don't have issues regarding deadlines. Deadlines can be met in any case. HW is powerful enough.

But maybe you can give some practical advice, because this is really what I don't understand properly.
What is better for FAH:
- starting calculating a WU for ~3hrs, then pause it for ~12hrs and finish folding which takes another ~3hrs. So total processing time is ~18hrs instead of ~6hrs
or
- stop folding early, do not start folding for ~3h and start folding a new WU after the "12hrs pause period". Finish it in 6hrs but wasting 3hrs idle time.

I just want to contribute in the best way possible.

[BobWilliams757]

There are options, but not all of them might work for you. But before we even get to that point, what are you using the computer for that is so intensive you can't fold and do those tasks at the same time? We surf the 'net, stream videos, edit photos and videos, work on spreadsheets, etc.... all while never pausing folding. I've even run multiple benchmarks at the same time while folding, trying to find any possible instabilities. Even with the older hardware we are running, it's surprising how little the folding slows the system down.

But in any case, some options below:

Pause folding when you have to do computer intensive tasks, then resume when those intensive tasks are over with.

Since you mentioned long CPU tasks, lower the thread count on your slot. Usually the larger projects require more cores, and don't as often assign to lower core count. And even if they do, you have more free threads to run your other tasks and fold at the same time.

Fold primarily when you sleep. Most deadlines are generous enough to fold at night and pause during the day.

In the scenario you gave in your post above, I would just keep folding and do my other tasks at the same time. Instead of a potential 18 hour time, it might take 6.5 instead of the optimal 6.

[foldinghomealone]

It's like that:
After work, if I have time, I use the PC for gaming and therefore no CPU folding possible.
Due to lower energy costs, the night is reserved for MPS GPU folding as it has the highest power consumption. Parallel CPU folding drops PPD a lot / too much.
During day time I fold some CPU WUs, as the power consumption is lower but energy costs are higher.

[muziqaz]

As long as you return the WU within the timeout, you are good. FAH prefers quicker return rate. How you achieve that is up to you. That does not mean everyone should buy high end hardware for FAH. But the only thing we care is that WU comes back within timeout

[Joe_H]

After work, if I have time, I use the PC for gaming and therefore no CPU folding possible.

I game on my system, and leave CPU folding running. I don't see any slowdowns in games. That is using 6 of the 8 performance cores. You do not need to use all cores for CPU folding as has already been state. You can also use few CPU cores and leave plenty of CPU headroom while folding on your GPU. You mention PPD, are you folding for points or to complete WUs to contribute to the project?

There are options such as using fewer core threads, pausing while gaming, and so on. So post details of what you have for a CPU and GPU and more suggestions may come.

Getting back to your question about incomplete WUs being returned to be passed on to someone else. The F@h project did try that a number of years ago, the so-called streaming core. That experiment was ended, paraphrasing what I heard it was because it was a logistical mess. The same occurred when a server bug caused some projects to be issued with the wrong number of steps, the researchers tried to salvage data from the returned WUs, but in most cases ended up having to restart the projects.

WUs are set to run a set number of steps. When completed the returned WU is verified and then used to generate the next Gen WU that starts where the previous Gen ended. After a set number of Gens then that particular Run, Clone trajectory ends unless the researcher finds they need to extend it further.

A dumped WU at least generates a report to the server that the WU is not being completed, that ends the countdown to the Timeout and queues that particular Run, Clone and Gen for reassignment.

[foldinghomealone]

Gaming and folding at the same time is a no go, at least for me.

I care about both PPD and WU. That's why I'm asking those questions and wonder why it is done like it is done. And how I can contribute best.
And: Lower PPD mean also fewer WU/day. To a lesser extend but anyway...
When folding in Linux, I use MPS. It gets me ~30% more WUs per time but similar PPD compared to "normal" folding.

I have my HW config shown in the profile: Ryzen 5900X and RTX 5080