Welcome to MilkyWay@home

Problem Getting Multiple Tasks per Scheduler Reauest

Message boards : Number crunching : Problem Getting Multiple Tasks per Scheduler Reauest
Message board moderation

To post messages, you must log in.

AuthorMessage
Cherokee150

Send message
Joined: 17 Nov 14
Posts: 6
Credit: 14,036,355
RAC: 0
Message 70888 - Posted: 20 Jun 2021, 9:51:05 UTC

One of my computers, 879281, can receive up to 125 GPU work units per scheduler request. My other two computers, 879005 and 596332, are never able to get more than 6 tasks per request.

Does anyone know why this is? How can I raise the number of units per request on my two "6 per" computers?

Also, is there a reason why I can only get 300 GPU units per day for my 879281 and 879005 computers? My 879005 and my 879281 computer can process hundreds of GPU units per day. Is there a way to raise this limit? I don’t want them to sit idle.

Oh, and one final question. Why are there never any output files returned to Milky Way when units complete? Without receiving output from us, what does our processing accomplish for their project?

Thank you very much for any help you can give me with these questions.
ID: 70888 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 8 May 09
Posts: 3315
Credit: 519,950,525
RAC: 21,840
Message 70889 - Posted: 20 Jun 2021, 10:46:56 UTC - in response to Message 70888.  

One of my computers, 879281, can receive up to 125 GPU work units per scheduler request. My other two computers, 879005 and 596332, are never able to get more than 6 tasks per request.

Does anyone know why this is? How can I raise the number of units per request on my two "6 per" computers?


Check the cache size on the '6 per' pc's.

Also, is there a reason why I can only get 300 GPU units per day for my 879281 and 879005 computers? My 879005 and my 879281 computer can process hundreds of GPU units per day. Is there a way to raise this limit? I don’t want them to sit idle.


A choice by MilkyWay so every0one can crunch some instead of a few people crunching all.

Thank you very much for any help you can give me with these questions.[/quote]

No clue about the other questions.
ID: 70889 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cherokee150

Send message
Joined: 17 Nov 14
Posts: 6
Credit: 14,036,355
RAC: 0
Message 70892 - Posted: 21 Jun 2021, 12:17:02 UTC - in response to Message 70889.  

I have looked through all the preferences, but I don't see any caches that would appear to impact how many units would be sent per scheduler request. Can you specify to which cache size parameter you are referring, and in which XML file it resides?

Perhaps an example of the different way MilkyWay handles the PCs will help.

Due to circumstances, only one PC at a time can access the Internet, so I must swap the connection as needed.

1. I swap the Internet to PC 879281, which has completed 300 units. The PC reports those 300 units, then requests new units. MilkyWay sends 125 new units. About 95 seconds later the PC makes a second request for units. MilkyWay sends another 125 units. About 95 seconds later a third request gets 50 units, completing the replacement of the 300 completed units. Total elapsed time is just a little more than 3 minutes.

2. I swap the Internet to PC 879005. It has also completed 300 units. The PC reports those 300 units, then requests new units. MilkyWay sends 6 new units. About 95 seconds later the PC makes a second request for units. MilkyWay sends another 6 units. This cycle repeats about 50 times until the 300 units have been replaced. Total elapsed time is over 1 hour and 20 minutes.

Can you see what is causing this behavior and, hopefully, a way to fix it?

Thanks!
ID: 70892 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Keith Myers
Avatar

Send message
Joined: 24 Jan 11
Posts: 696
Credit: 540,067,113
RAC: 86,702
Message 70895 - Posted: 21 Jun 2021, 18:55:20 UTC - in response to Message 70892.  

Are you sure you have not limited the amount of memory or hard drive space allowed to Milkyway on the slow computer? It almost sounds like it has to crunch six tasks to make room for six new tasks.

The only time I have seen a similar problem of not getting many tasks upon request was at Seti and discovered the root cause was too little disk space allowed. As soon as I increased it, boom, the next request filled to the max from the server task buffer.
ID: 70895 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 8 May 09
Posts: 3315
Credit: 519,950,525
RAC: 21,840
Message 70896 - Posted: 22 Jun 2021, 10:34:17 UTC - in response to Message 70892.  

I have looked through all the preferences, but I don't see any caches that would appear to impact how many units would be sent per scheduler request. Can you specify to which cache size parameter you are referring, and in which XML file it resides?



2. I swap the Internet to PC 879005. It has also completed 300 units. The PC reports those 300 units, then requests new units. MilkyWay sends 6 new units. About 95 seconds later the PC makes a second request for units. MilkyWay sends another 6 units. This cycle repeats about 50 times until the 300 units have been replaced. Total elapsed time is over 1 hour and 20 minutes.

Can you see what is causing this behavior and, hopefully, a way to fix it?

Thanks!


Since you finally get the 300 units your cache size is not the problem, the cache setting on both on the website and on your pc, both are the same but the website settings affect every pc while if you do it thru Boinc Manager only that pc is affected and it is the setting that says "store at least [] days of work" and "store [] additional days of work.
ID: 70896 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cherokee150

Send message
Joined: 17 Nov 14
Posts: 6
Credit: 14,036,355
RAC: 0
Message 70905 - Posted: 23 Jun 2021, 9:13:00 UTC - in response to Message 70896.  

Thank you very much for explaining the reason I can't get more than 300 units per day.

Keith found the answer to my "6 per" problem. It was due to disk limits that I set in BOINC Manager years ago when space was much more of an issue for everyone. I updated those to values appropriate for today.

I am still trying to understand why output files are never sent to MilkyWay. If MilkyWay doesn’t get output from us, how does our processing help them? Do you, or does anyone, know how our results get back to MilkyWay?

Thanks!
ID: 70905 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cherokee150

Send message
Joined: 17 Nov 14
Posts: 6
Credit: 14,036,355
RAC: 0
Message 70906 - Posted: 23 Jun 2021, 10:14:40 UTC - in response to Message 70895.  

Keith, you found the answer to my "6 units per request" problem!

I set up BOINC Manager's preferences for each computer years ago, when disk drives didn't have so much space. It was so long ago that I didn't think of it when I was trying to debug this. With your help, I updated the "Disk and memory" preferences for each computer. The problem was immediately solved!

I am still wondering how the scientists at MilkyWay benefit from our processing, since no result files are ever sent back to them. Is there some other way they are able to get the units' results? Do you, or does anyone, know how this works in MilkyWay?

Thank you once again, Keith, for guiding me to the right path!
ID: 70906 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Keith Myers
Avatar

Send message
Joined: 24 Jan 11
Posts: 696
Credit: 540,067,113
RAC: 86,702
Message 70908 - Posted: 23 Jun 2021, 19:39:34 UTC - in response to Message 70906.  

MW does send notification to the project for every finished task. The output file is just the stderr.txt which is quite small and uploads instantly. The problem is that because of the way that MW orchestrates task reporting and task requesting, you don't see your upload back to the project until the client contacts the project and DOESN'T ask for work. So the uploads occur in batches.

You can prove it to yourself by setting the Event Log option for http_xfer_debug and read the logfile output where you can see the output file going upstream to the project.

If you look on the project for your finished task results, you can see how the task was run and on which device and the parameter configuration. And remember each task is a bundled set of 4 or 5 individual tasks.
ID: 70908 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cherokee150

Send message
Joined: 17 Nov 14
Posts: 6
Credit: 14,036,355
RAC: 0
Message 70915 - Posted: 25 Jun 2021, 9:03:07 UTC - in response to Message 70908.  

Thank you very much, Keith. Now I understand how it works.
ID: 70915 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Number crunching : Problem Getting Multiple Tasks per Scheduler Reauest

©2024 Astroinformatics Group