Invalid tasks
Message boards :
Number crunching :
Invalid tasks
Message board moderation
Author | Message |
---|---|
Send message Joined: 8 Apr 15 Posts: 34 Credit: 389,238 RAC: 0 |
Noticed that as of this post I have 54 "Invalid" tasks across my various PC's spread over many days of computation. My question is, what makes them "invalid"? I am not having any invalid tasks at any other projects. |
Send message Joined: 6 Mar 23 Posts: 36 Credit: 2,078,354 RAC: 0 |
Noticed that as of this post I have 54 "Invalid" tasks across my various PC's spread over many days of computation. I have about 10 Invalid tasks at the moment. I looked up the work units involved to see if there were others with the same work unit, and how they did. In summary, in each case there were three users (including me) who got each of these work units. Mine was invalid and the other two were successful. As far as I can tell we all got the work unit about the same time, and usually the other two users got theirs slightly earlier than mine.But not always. But since they returned their results first, my job was canceled by the server. I am not sure if the server should have sent me the task at all until one of the others failed or timed out. But they did. |
Send message Joined: 14 Apr 22 Posts: 11 Credit: 10,554,385 RAC: 0 |
I currently have 126 invalid tasks which is more than I'm used to seeing with other projects. This project seems to have more invalids than other projects but mine is less than 1% of the total number computed. Some projects feel that is acceptable. |
Send message Joined: 31 Mar 18 Posts: 3 Credit: 10,172,753 RAC: 0 |
Looking at the returned results for example workunit # 5884087 https://denis.usj.es/denisathome/workunit.php?wuid=5884087, there are three invalids and two valids. Probing into the stderr output file for each task ID, I'm guessing that for validation to be successful the output result (looks like there are 44 parameters or perhaps more) must match with a given same input parameters (guessing there are 18 parameters). I don't know what is causing the other three PCs to return different values in the output results Not claiming that I'm an expert in this but just an observation. Perhaps the admin can shed some light into this. |
Send message Joined: 18 Mar 15 Posts: 284 Credit: 2,748,608 RAC: 0 |
Hi all, The percentage of invalid task is very small. The validation is not done with the output result, is done with the results files (out and markers). It compares both files of the results with other user. Small differences could appear due to different implementations of the opperations in different processors. The validation uses an error margin, but if the differences are bigger, a new simulation is requested. Some differences could be solved or reduced with the implementation and we work on analyze that. At begining of this restart of the project, we had some problems with the checkpoints, but that is now solved. We will continue working to try to reduce it further. Best, Jesús. Jesús Carro Universidad San Jorge @InSilicoHeart |
Send message Joined: 14 Apr 22 Posts: 11 Credit: 10,554,385 RAC: 0 |
Most of my invalids look like my Linux machine was being compared to the output of a Windows machine and my result was marked invalid when the result from another Windows machine was returned. Maybe the project should consider implementing Homogeneous Redundancy. Looking at the last 1000 WUs just sent to me, it looks there is a considerable resend burden on the server. A significant percentage of the work just sent is resends. |
Send message Joined: 24 Jul 23 Posts: 6 Credit: 105,097 RAC: 0 |
Hello, I am also getting some occasional invalid tasks. Example NHuVe_20231106172652884480_NHuVeNLOnlyG_k_21-Schmidt_500-conf_465 Am I correct that this is not an issue with my machine, there is nothing I need to change and I can continue to process tasks? It's only occasional so it's not causing me a problem, I just want to check that my machine is not the cause of the problem. Thanks |
Send message Joined: 18 Mar 15 Posts: 284 Credit: 2,748,608 RAC: 0 |
No, most of the invalid causes are problems due to differences in the OS we have no considered. We need to still improve the program to avoid them. Best, Jesús Jesús Carro Universidad San Jorge @InSilicoHeart |
Send message Joined: 24 Jul 23 Posts: 6 Credit: 105,097 RAC: 0 |
Thanks for your reply. I will continue to process tasks. |
Send message Joined: 11 Oct 23 Posts: 25 Credit: 3,978,292 RAC: 0 |
No, most of the invalid causes are problems due to differences in the OS we have no considered. We need to still improve the program to avoid them.Any news on this front? I get (relatively speaking) a lot of invalid results too, as well as "inconclusive" ones. Certainly at a much higher percentage than on any other project I am crunching for. All hosts on my side running DENNIS are Windows 10, Windows Server 2016/2019, one MacBook Pro and a couple of Windows 7 hosts still... Ralf |
Send message Joined: 6 Mar 24 Posts: 1 Credit: 175,373 RAC: 0 |
I am also seeing a high number of invalid tasks. I am almost at 25% of all tasks being invalid. Not sure why this is occurring. I am running 10 separate Windows 10 Pro machines on Intel i5-6500 CPU. It appears if I don't manually request updates then the tasks complete successfully. However if I have tasks ready to report and I request an update, they change to Invalid before being removed from the computers. Not sure if this helps. Thanks for letting me participate. |
Send message Joined: 29 Feb 24 Posts: 9 Credit: 299,776 RAC: 0 |
TPCBF, you might want to look into using 'app_config.xml' to limit your number of cores on your Intel® Core™ i7-12700 Processor computer. Maybe try limiting to 8 cores or less since that is what the Intel specs show as 'performance cores' I am running a Ryzen 1700 with SMT turned off so 8 cores total and process tasks in under an hour. I am not showing errors when I last checked. I think this code below would do it. Exit BOINC manager and stop all tasks. You would open Notepad and paste it in. Save it as 'app_config.xml' making sure to 'Save as type: all files'. You would save it in the BOINC project directory for DENNIS. Something like C:\ProgramData\BOINC\projects\denis.usj.es_denisathome <?xml version="1.0" encoding="utf-8"?> <!-- This is a comment --> <app_config> <project_max_concurrent>8</project_max_concurrent> </app_config> |