Hello there,

i have the following problem: I've got a big bunch of astropulse workunits scheduled for intel GPU (Server sees them as CPU) which are way to much and the intel GPU will be not able to crunch all of them on time. I already stopped BOINC to fetch AP units and think now of rescheduling the surplus AP units from intel GPU to Nvidia GPU. Unfortunately i found no way to reschedule between GPUs, only from CPU to GPU and vice versa. Any advice here, because i don't want to trash the WUs thereby annoying my wingmen?

---

Aloha, Uli

Personally I wouldn't do anything, just let the system work it out. Any that you can't complete in time will be sent out again to others.

Cheers.

Unfortunately i found no way to reschedule between GPUs, only from CPU to GPU and vice versa

I think you can do this in 2 steps:
Intel GPU -> CPU
CPU -> Nvidia GPU

(by 'Preferred use:' - I think you'll need existing CPU AP app in app_info.xml
After adding CPU AP app to app_info.xml you have to restart BOINC so new info from app_info.xml is entered by BOINC to client_state.xml

First use 'Simulation mode, only create a new state file' to see/compare the changes in client_state.xml
)

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

Personally I wouldn't do anything, just let the system work it out. Any that you can't complete in time will be sent out again to others.

That's the best way to go, if you are sure you can't cruch them all, you can eventually abort some, so they can be send out sooner. But actually you don't need to do anything, BOINC is designed to handle such thing on it's own.

---

Thank you all for your suggestions.
Greatly appreciated!

---

Aloha, Uli

Hello there,

i have the following problem: I've got a big bunch of astropulse workunits scheduled for intel GPU (Server sees them as CPU) which are way to much and the intel GPU will be not able to crunch all of them on time. I already stopped BOINC to fetch AP units and think now of rescheduling the surplus AP units from intel GPU to Nvidia GPU. Unfortunately i found no way to reschedule between GPUs, only from CPU to GPU and vice versa. Any advice here, because i don't want to trash the WUs thereby annoying my wingmen?

If you aren't over your GPU limit of 100, you can set your preferences for Use NVIDIA GPU only and then remove the wanted files from the client_state.xml. They will be resent as lost tasks to the device your preferences indicate. Anything over the limit will be listed as Timed out - no response and sent to others.

First, set the preferences to Use Only NVIDIA GPU and then hit the Update button a few times until it is asking for NVIDIA. Select the AP files you want to transfer and Suspend them. Shut down BOINC and wait about 15 seconds. Open the client_state.xml with notepad and search for suspended. Remove the entire results entry for each Suspended file, everything between and including <result> ... </result>. Save the client_state.xml. Don't make any mistakes or the entire file will be rejected and everything resent. Launch BOINC, then hit the Update button until all Lost Tasks are resent in groups of 20. Then adjust your preferences to normal.

Uli, I don't leave anything up to Boinc as it tends at times to over fetch and can't complete things on time.
I just suspend the currently running ones and let the others run up to a few percent. That way you can complete them and the server is not any wiser, as it thinks you started them on time.

---

Pluto will always be a planet to me.

Seti Ambassador
Not to late to order an Anni Shirt

Its not starting on time but completing on time.

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Sorry Rob. but it has worked for me.
The server will only cancel not started on time. So if you get a few percent in, you are fine.
Rob did you do a test run like I did?

---

Pluto will always be a planet to me.

Seti Ambassador
Not to late to order an Anni Shirt

yes, I had a few on a retiring cruncher that timed out, not started on time, and they were aborted by the server.

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

The only thing is that doing that is what winds up getting some w/u's stuck in limbo.

Cheers.

They timed out, because they were not started in time.
Thus my fiddle.

---

Pluto will always be a planet to me.

Seti Ambassador
Not to late to order an Anni Shirt

The only thing is that doing that is what winds up getting some w/u's stuck in limbo.

Cheers.

Sorry Wiggo, I only have one stuck.
It is part of a known problem.

I was born with black hair, then lost it all and returned as a blond with curls.
Now I am a Brunette with some gray.

---

Pluto will always be a planet to me.

Seti Ambassador
Not to late to order an Anni Shirt

I was born with black hair, then lost it all and returned as a blond with curls.
Now I am a Brunette with some gray.

Myself I'm grey with a little brunette. ;-)

Cheers.

The only thing is that doing that is what winds up getting some w/u's stuck in limbo.

Cheers.

Sorry Wiggo, I only have one stuck.
It is part of a known problem.

Yes, and since we know the problem, we also know what to do for it should not occur more often than really unavoidable: abort all WUs which we can't finish in time, i.e. before the replacement result is send out. Most of the resends will be crunched successfully, so crunching timed out WUs is not only causing a problem on the server side, but is just a plain waste of own resources and definitely nothing that should be recommended IMHO.

---

I'm sorry Uli but as much as I look I just can't see your stuck 1 in your task list.

Can you give us a link to it?

Cheers.

http://setiathome.berkeley.edu/workunit.php?wuid=1265039314
Happy to help Wiggo.
Just part of the known problem.

---

Pluto will always be a planet to me.

Seti Ambassador
Not to late to order an Anni Shirt

If you aren't over your GPU limit of 100, you can set your preferences for Use NVIDIA GPU only and then remove the wanted files from the client_state.xml. (...)

Thanks a lot, that worked and the resent files are also correctly marked as Nvidia! :)

---

Aloha, Uli

http://setiathome.berkeley.edu/workunit.php?wuid=1265039314
Happy to help Wiggo.
Just part of the known problem.

Ah yes, that was when that script was run on the servers.

We maybe waiting a while on those to be corrected Uli (hopefully at the same time the old V6 Enhanced database gets cleaned out).

Cheers