Problems and Technical Issues with Rosetta@home

Message boards : Number crunching : Problems and Technical Issues with Rosetta@home

To post messages, you must log in.

Previous · 1 . . . 37 · 38 · 39 · 40 · 41 · 42 · 43 . . . 55 · Next

AuthorMessage
Profile David E K
Volunteer moderator
Project administrator
Project developer
Project scientist

Send message
Joined: 1 Jul 05
Posts: 1018
Credit: 4,334,829
RAC: 0
Message 77317 - Posted: 10 Aug 2014, 5:18:08 UTC

I fired up more make_work daemons to hopefully catch up with the work demand. We have plenty of work queued up but our daemons were having trouble catching up. Hopefully the updates I just made will help.

Sorry but I had to stop the servers and restart so there was a short bit of down time, so short I didn't bother posting anything.
ID: 77317 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Terminal*

Send message
Joined: 23 Nov 05
Posts: 6
Credit: 7,845,878
RAC: 0
Message 77319 - Posted: 10 Aug 2014, 6:31:17 UTC
Last modified: 10 Aug 2014, 6:44:06 UTC

I have a few 40 core machines i'm trying to get work on, and they just keep getting refused new work :(

Do you guys take local hardware donations? I could part with several Poweredge 2950's.

Most of the links on https://boinc.bakerlab.org/rosetta/rah_donations.php are broken links.
ID: 77319 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Greg_BE
Avatar

Send message
Joined: 30 May 06
Posts: 5658
Credit: 5,633,150
RAC: 945
Message 77321 - Posted: 10 Aug 2014, 14:53:26 UTC - in response to Message 77317.  

Oh! Well that is good news then. I guess it was just a bit of bad timing on our end here in Europe. Here's to more smooth running in the future.

I fired up more make_work daemons to hopefully catch up with the work demand. We have plenty of work queued up but our daemons were having trouble catching up. Hopefully the updates I just made will help.

Sorry but I had to stop the servers and restart so there was a short bit of down time, so short I didn't bother posting anything.

ID: 77321 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Polian
Avatar

Send message
Joined: 21 Sep 05
Posts: 152
Credit: 10,141,266
RAC: 0
Message 77322 - Posted: 10 Aug 2014, 18:07:58 UTC

My pending results are starting to spool up... You poor guys, fix one bottleneck and the next appears :) I suppose it's a good problem to have though, with all the new users.
ID: 77322 · Rating: 0 · rate: Rate + / Rate - Report as offensive
googloo
Avatar

Send message
Joined: 15 Sep 06
Posts: 133
Credit: 21,531,321
RAC: 6,547
Message 77323 - Posted: 11 Aug 2014, 11:17:23 UTC

I'm running out of Rosetta tasks again.

Ready to send 22
ID: 77323 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Sid Celery

Send message
Joined: 11 Feb 08
Posts: 1960
Credit: 38,076,311
RAC: 6,958
Message 77324 - Posted: 11 Aug 2014, 13:14:46 UTC
Last modified: 11 Aug 2014, 13:17:07 UTC

Just mopping up the last couple of the 55 WCG tasks I had to grab on Sunday morning when the scheduler was switched off.

I've returned 15 Rosetta tasks already since midnight, but only 1 has validated. The other 14 are still showing as pending - some from 15 hours ago.

The validators are showing as running on the server status page.

Edit: And struggling to get tasks, just as googloo reports. 2726 ready to send. In progress very low at 305k compared to a high of up to 2-3m during the last week
ID: 77324 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Sid Celery

Send message
Joined: 11 Feb 08
Posts: 1960
Credit: 38,076,311
RAC: 6,958
Message 77325 - Posted: 11 Aug 2014, 17:16:58 UTC

Tasks now coming down and all delayed validation up to date.

Thanks all
ID: 77325 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile David E K
Volunteer moderator
Project administrator
Project developer
Project scientist

Send message
Joined: 1 Jul 05
Posts: 1018
Credit: 4,334,829
RAC: 0
Message 77326 - Posted: 11 Aug 2014, 18:24:00 UTC

Is anyone still having issues getting work?
ID: 77326 · Rating: 0 · rate: Rate + / Rate - Report as offensive
googloo
Avatar

Send message
Joined: 15 Sep 06
Posts: 133
Credit: 21,531,321
RAC: 6,547
Message 77327 - Posted: 11 Aug 2014, 18:43:08 UTC - in response to Message 77326.  

Is anyone still having issues getting work?


Better now, thanks.
8/11/2014 12:28:31 PM | rosetta@home | Scheduler request completed: got 20 new tasks
ID: 77327 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Sid Celery

Send message
Joined: 11 Feb 08
Posts: 1960
Credit: 38,076,311
RAC: 6,958
Message 77328 - Posted: 11 Aug 2014, 23:00:58 UTC - in response to Message 77326.  

Is anyone still having issues getting work?

I thought I was, but I think this is the definition of no:

11/08/2014 23:50:48 | rosetta@home | update requested by user
11/08/2014 23:50:51 | rosetta@home | Sending scheduler request: Requested by user.
11/08/2014 23:50:51 | rosetta@home | Not requesting tasks: don't need


ID: 77328 · Rating: 0 · rate: Rate + / Rate - Report as offensive
TPCBF

Send message
Joined: 29 Nov 10
Posts: 109
Credit: 4,609,497
RAC: 1,614
Message 77329 - Posted: 12 Aug 2014, 0:35:39 UTC - in response to Message 77326.  

Is anyone still having issues getting work?
No, now I am getting new WUs, more than before, but for some strange reasons, SETI@HOME,which "shares" the CPU cycles on that machine "isn't getting any"... :-(
ID: 77329 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Mod.Sense
Volunteer moderator

Send message
Joined: 22 Aug 06
Posts: 4018
Credit: 0
RAC: 0
Message 77330 - Posted: 12 Aug 2014, 2:43:12 UTC - in response to Message 77329.  

Is anyone still having issues getting work?
No, now I am getting new WUs, more than before, but for some strange reasons, SETI@HOME,which "shares" the CPU cycles on that machine "isn't getting any"... :-(


This is normal. BOINC Manager is trying to maintain your defined resource share between the two projects. When your machine went without R@h work for a period of time your machine was completing more work for SETI than your resource share would normally have allowed for SETI. And so SETI got ahead of schedule (where "schedule" is relative to your defined resource shares). So, to reestablish the balance between the two, now the BOINC Manager will request additional work from R@h to even things out.

Since it was only a day or two of getting behind, it should only take a day or two for the two projects to reach parity again and for you to get back to a stream of work in proportion to the configured resource shares.
Rosetta Moderator: Mod.Sense
ID: 77330 · Rating: 0 · rate: Rate + / Rate - Report as offensive
neil

Send message
Joined: 22 Dec 06
Posts: 3
Credit: 18,133,363
RAC: 35
Message 77406 - Posted: 25 Aug 2014, 16:33:32 UTC

Greetings:

I'm not convinced this is 'normal'. My PC runs Rosetta and World Community Grid (WCG) currently. If I set Rosetta to 'No new tasks' mode and let the system catch up, meaning compute all of the Rosetta work, then let the PC run nothing but WCG for a week or two, all I have to do is set Rosetta to 'Allow new tasks' and let it run for 24 to 48 hours before the system is swamped with Rosetta tasks and WCG is getting all but completely shut out.

The Rosetta jobs all start 'Running, high priority' as the system comes to them and move to the top of the queue.

I don't think Rosetta plays well with other projects. If I had to guess, not being extremely knowledgeable of how the Boinc algorithm works, I'd say it's because of the short deadlines assigned to work units.

The majority of the Rosetta and WCG work units have similar estimated run-times.

As it now stands on my system, I enabled Rosetta again yesterday, I have 175 Rosetta units and 64 WCG units.

Maybe I'm all wet, but it does appear that Rosetta is much more aggressive in it's share of resources than WCG, in this case (I've run other projects with it as well).

I'm getting close to removing Rosetta from my system because of this.

My two cents...

ID: 77406 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Miklos M

Send message
Joined: 8 Dec 13
Posts: 29
Credit: 5,277,251
RAC: 0
Message 77407 - Posted: 25 Aug 2014, 16:53:57 UTC

I am getting the extra long timed wu's and a couple, so far resulted in errors. I did not abort any units, at least not lately. My computer is set for one day time limit per unit, but these units seem to need up to 50 hours per cpu.
ID: 77407 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Murasaki
Avatar

Send message
Joined: 20 Apr 06
Posts: 303
Credit: 511,418
RAC: 0
Message 77408 - Posted: 25 Aug 2014, 17:59:35 UTC - in response to Message 77406.  
Last modified: 25 Aug 2014, 18:16:44 UTC

Greetings:

I'm not convinced this is 'normal'. My PC runs Rosetta and World Community Grid (WCG) currently. If I set Rosetta to 'No new tasks' mode and let the system catch up, meaning compute all of the Rosetta work, then let the PC run nothing but WCG for a week or two, all I have to do is set Rosetta to 'Allow new tasks' and let it run for 24 to 48 hours before the system is swamped with Rosetta tasks and WCG is getting all but completely shut out.


Yep, sounds normal to me. From your description the debt imbalance between WCG and Rosetta is skewed. You may have had a time when your system was only able to do WCG tasks (for example, during the Rosetta server strain at the start of August) and your client is now trying to compensate by downloading more Rosetta work.

Depending on how BOINC is interpreting the "no new tasks" command you are just maintaining the existing problem (freeze the comparative debt levels between the two projects) or making the situation worse (Rosetta's debt increases while tasks are frozen).


I don't think Rosetta plays well with other projects. If I had to guess, not being extremely knowledgeable of how the Boinc algorithm works, I'd say it's because of the short deadlines assigned to work units.


I run Rosetta and WCG too and don't see any of the problems you describe. In fact, Rosetta usually gives a week longer than WCG for task deadlines.


I'd suggest setting "no new tasks" for both projects to finish all existing work then remove both projects and reattach. That should clear any debt issues for both projects.

Alternative solutions would be to:

  • detach just one project and reattach (will only work if the underlying issue is with that particular project).
  • locate the long term debt setting in your BOINC data files and reset it (I can only find instructions do this for BOINC 6.x).
  • let both projects run without interference. Eventually the debt will come back into balance and normal behaviour will resume.



There are other issues that could be causing some of the symptoms you describe. For example, if you have a large work buffer set in your computing preferences your client may be asking Rosetta to fill it to the maximum to help you catch up. If your computer isn't quick enough to clear the buffer that would explain the high priority work, as tasks start to fail. Having a proxy server cache tasks for you would have a similar impact.



Here is a reply to a similar issue a few years ago:

Greg_BE wrote:

yep, that happened to me as well.
Rosie ran out of work and Einstein and Poem loaded me up big time.
Then Rosie came back online and Boinc manager had to sing high priority for awhile while clearing the backlog of the other projects.

Then it started loading Rosie stuff in and not the others due to debt levels and finally settled down a week or 2 after that. If you watch your 'user average' stats chart, you will see that one of your projects is way higher than Rosie in credit and then you should see that project drop off in work a bit and Rosie loading in more work to get your credit back up to where it was.

But like we said, just sit back, let things run and Boinc manager will sort it all out for you over time and then return to normal.
ID: 77408 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Mod.Sense
Volunteer moderator

Send message
Joined: 22 Aug 06
Posts: 4018
Credit: 0
RAC: 0
Message 77410 - Posted: 25 Aug 2014, 19:03:46 UTC - in response to Message 77406.  

Maybe I'm all wet, but it does appear that Rosetta is much more aggressive in it's share of resources than WCG, in this case (I've run other projects with it as well).


As described by others, there are many factors that go in to which projects the BOINC Manager chooses to run and to request work from. But it's important to keep in mind that it is the BOINC Manager that "decides", not the projects. It is trying to optimize what can be conflicting goals. The only thing I can think of that might cause it to look like R@h is overtaking other projects is that some R@h tasks do have an extended time (over an hour) between checkpoints. In order to preserve and maximize the work your machine produces, the BOINC Manager will sometimes wait to interrupt a task until it reaches a checkpoint. But this would not cause an effect over the course of days.

Another possibility is that R@h tasks use significantly more memory than many other projects. Depending on how your BOINC Manager is configured for the amount of memory it is allowed to use, it might get knee deep in to a R@h tasks that needs more memory to complete and in order to make that possible the BOINC Manager might hold other tasks until the memory usage goes back down.

The BOINC Manager also can play by different rules when a project is marked for no new work, when tasks are suspended, when it's estimates of runtimes are off, and etc.

I'm certainly *NOT* calling you all wet, but please don't make the current observations be the reason for withdrawing your support of the project. There is a moderator (the BOINC Manager) between the project and getting machine resources. It is certainly possible it is not doing the best job at it, but more likely just that it is looking at the debts and etc. and actually trying to enforce the resource shares that you set up between the projects.
Rosetta Moderator: Mod.Sense
ID: 77410 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Miklos M

Send message
Joined: 8 Dec 13
Posts: 29
Credit: 5,277,251
RAC: 0
Message 77411 - Posted: 25 Aug 2014, 21:33:35 UTC

683255938 619208054 15 Aug 2014 7:50:59 UTC 25 Aug 2014 17:55:37 UTC Over Validate error Done 67,088.14 --- ---
The errors keep coming. Perhaps time to delete those new 48+ hour wu's.
ID: 77411 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Murasaki
Avatar

Send message
Joined: 20 Apr 06
Posts: 303
Credit: 511,418
RAC: 0
Message 77413 - Posted: 25 Aug 2014, 22:06:24 UTC - in response to Message 77411.  
Last modified: 25 Aug 2014, 23:01:45 UTC

683255938 619208054 15 Aug 2014 7:50:59 UTC 25 Aug 2014 17:55:37 UTC Over Validate error Done 67,088.14 --- ---
The errors keep coming. Perhaps time to delete those new 48+ hour wu's.


You seem to have a mix of units with different types of errors. I have reported your issues with long file names on the Minirosetta 3.52 thread.

You may want to report the different types of issues you are seeing there.

Edit: I see you mentioned the long task name already a couple of posts above me in that thread. Looking through your recent tasks there are only a few validate/compute errors caused by the system that I can see. However, a large number of errors were reported as tasks aborted by the user. I expect it will be useful to explain further why you aborted them. Were they not checkpointing? Were they all running at x% longer than your preferred run time as you mentioned above?
ID: 77413 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Miklos M

Send message
Joined: 8 Dec 13
Posts: 29
Credit: 5,277,251
RAC: 0
Message 77416 - Posted: 26 Aug 2014, 17:25:39 UTC - in response to Message 77413.  

I aborted them last night since it seems all the 48+hour ones were in some way ending in errors. The usual 24-25 hour ones have no problems. I am waiting until Rosetta sends the latter ones again.
683255938 619208054 15 Aug 2014 7:50:59 UTC 25 Aug 2014 17:55:37 UTC Over Validate error Done 67,088.14 --- ---
The errors keep coming. Perhaps time to delete those new 48+ hour wu's.


You seem to have a mix of units with different types of errors. I have reported your issues with long file names on the Minirosetta 3.52 thread.

You may want to report the different types of issues you are seeing there.

Edit: I see you mentioned the long task name already a couple of posts above me in that thread. Looking through your recent tasks there are only a few validate/compute errors caused by the system that I can see. However, a large number of errors were reported as tasks aborted by the user. I expect it will be useful to explain further why you aborted them. Were they not checkpointing? Were they all running at x% longer than your preferred run time as you mentioned above?

ID: 77416 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile David E K
Volunteer moderator
Project administrator
Project developer
Project scientist

Send message
Joined: 1 Jul 05
Posts: 1018
Credit: 4,334,829
RAC: 0
Message 77417 - Posted: 26 Aug 2014, 18:02:32 UTC - in response to Message 77407.  

I am getting the extra long timed wu's and a couple, so far resulted in errors. I did not abort any units, at least not lately. My computer is set for one day time limit per unit, but these units seem to need up to 50 hours per cpu.



What are the names of these workunits? Sorry if you've mentioned them before below. I just can't find them. Thanks!

If there are bad work units that you believe are general errors resulting from a bug, please feel free to email me directly with the name and id of the work unit, the name is enough though. dekim at u dot washington dot edu



ID: 77417 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Previous · 1 . . . 37 · 38 · 39 · 40 · 41 · 42 · 43 . . . 55 · Next

Message boards : Number crunching : Problems and Technical Issues with Rosetta@home



©2024 University of Washington
https://www.bakerlab.org