(This was posted in Windows Q & A and I now realize that that is not a good place to put it as it probably has nothing to do with a Windows installation. I hope this location is better.)

I recently increased the number of days WUs I wanted, so as to weather outages here and elsewhere (mostly elsewhere). So I now have 4 rosetta WUs, one of which is ready to report. But despite differing deadlines, the three remaining ones are advancing to completion together. OK, only one is running at a time, but they are at 35%, 37% and 42% right now. I would expect the scheduler to run the one with the nearest Report deadline and when that completed, the next nearest, etc. But that's just me, I guess.

It would seem to me, if this pattern continues, that all three will complete around the same time and I'll be out of rosetta WUs, unless I get one or more right then. I am always connected to the internet (DSL) and I have requested enough additional work for 2 days.

I do have another project (SETI) and it's progressing as I would expect, one WU at a time. Is some setting of mine bogus?

Ross

Any clues in the BOINC message log? Any "waiting for memory" messages?
Rosetta WUs increase the amount of memory they need as they progress so it's possible for them to begin processing without problems but hit the limits set in your preferences sometime during the run. When this happens BOINC will stop processing that particular workunit and try its luck with another and I believe the result will look very much like what you describe. So my suggestion would be to review the percentage of memory you are allowing BOINC to use. (Within the first twenty lines or so of the BOINC message log you should see lines translating these percentages into MBs and if you want you can also check how much memory a particular task is using. On my Mac I use the Activity Monitor for this, on Windows I think the appropriate app is the Task Manager. Or you could just increase the percentage of memory BOINC is allowed to use and see how it responds.)

Snags

Might be a debt problem. Also, what are the resource share for each of your projects (SETI & Rosetta)?

Try this from the command line (in the BOINC bin directory):
boinccmd --set_debts https://boinc.bakerlab.org/rosetta 0 0

And let BOINC work itself out.

Might be a debt problem. Also, what are the resource share for each of your projects (SETI & Rosetta)?

Try this from the command line (in the BOINC bin directory):
boinccmd --set_debts https://boinc.bakerlab.org/rosetta 0 0

And let BOINC work itself out.

I could see this if it's a three cpu machine with four workunits, three rosetta and one SETI alternating for cpu time could look like this though I don't think it would necessarily indicate an actual problem with debt. But if it is a single cpu machine or BOINC is restricted to use only one cpu (by the "use at most x% of processors" preference) I don't see how debt issues would cause more than one workunit from the same project to be started.
So, Ross, how many cpus does this machine have and if more than one, what % of processors is BOINC allowed to use? And again, if you search back through your BOINC message log to the lines regarding preferences you'll see exactly how BOINC has interpreted this preference.

Snags

Are you expecting all CPU time to go to a single task until it is completed? Because this isn't generally how BOINC projects work. Generally they run one task PER CPU. But now I see your machine only has one CPU, so that's the same thing :)

If you actually do have 3 or 4 tasks all in a running state at the same time, this would indicate to me that you have altered the ncpus tag in your cc_config.xml file. This is generally used to test the BOINC client when you want to simulate more CPUs then your test machine actually has.

For general use, it is best not to specify this tag in the cc_config.xml file. This will then use the setting for the number of CPUs BOINC is allowed to use specified in your other preferences. And so in your case, it should then revert to only running a single task at a time. Less simultaneous tasks will help assure tasks do not conflict with each other for resources such as memory and swap space.

...and more directly to your point, the scheduler would notice the time to completion estimates dropping as the work progresses, and request more work in advance of the tasks reaching completion. So as to maintain the cache of work you have specified. It is certainly not an exact thing, many projects are not able to accurately estimate the time to completion of their tasks etc. but the BOINC client tries to keep you close to your desired backlog of work.

---

Rosetta Moderator: Mod.Sense

The only config.xml file I found related to BOINC is get_project_config.xml in the BOINC folder. I looked at it (for the first time) and found nothing interesting. I've looked at several other files in the same folder (C:Documents and SettingsAll UsersApplication DataBOINC) and in global_prefs.xml I found

<work_buf_additional_days>2</work_buf_additional_days>
<max_cpus>2</max_cpus>

I had thought I had requested 4 additional days and I have no idea where anyone got the idea I have 2 cpus.

The Belarc Advisor says I have a 2.20 gigahertz AMD Athlon 64. I'm sure it wasn't blazing hot even when new.

Aha! In my rosetta preferences, the question

"On multiprocessors, use at most 2 processors" may be the problem.

I have changed it to 1 processors, but the phrasing is misleading at best, a bug at worst.

BTW, each of the two projects has a 50 share. And I WISH I had three processors to play around with, but they probably wouldn't do me any good on a Windows machine.

Ross

---

From my messages:

3/12/2010 4:08:17 PM rosetta@home update requested by user
3/12/2010 4:08:21 PM rosetta@home Sending scheduler request: Requested by user.
3/12/2010 4:08:21 PM rosetta@home Not reporting or requesting tasks
3/12/2010 4:08:26 PM rosetta@home Scheduler request completed
3/12/2010 4:08:26 PM rosetta@home General prefs: from rosetta@home (last modified 12-Mar-2010 16:08:06)
3/12/2010 4:08:26 PM rosetta@home Computer location: home
3/12/2010 4:08:26 PM rosetta@home General prefs: no separate prefs for home; using your defaults
3/12/2010 4:08:26 PM Preferences:
3/12/2010 4:08:26 PM max memory usage when active: 1247.24MB
3/12/2010 4:08:26 PM max memory usage when idle: 2245.03MB
3/12/2010 4:08:26 PM max disk usage: 23.13GB
3/12/2010 4:08:26 PM don't use GPU while active
3/12/2010 4:08:26 PM suspend work if non-BOINC CPU load exceeds 25 %
3/12/2010 4:08:26 PM (to change, visit the web site of an attached project,
3/12/2010 4:08:26 PM or click on Preferences)

No mention of number of processors or number of days WU to keep ahead. (sigh)

Ross

Don't confuse "use at most" with any belief that you HAVE a given number of processors.

Go to the advanced view, to the tasks tab, are all of the tasks you describe in a status of "running" at the same time?

...so you are saying the cc_config.xml file does not exist in the data directory on your machine?

Now I'm thinking that the tasks are not all actually running. They are begun and probably hitting a limit on the amount of memory you are allowing BOINC to use, and then being suspended and will show a status of "waiting for memory". Another task is then begun, runs for a while and then also exceeds the memory you are allowing BOINC to use.

By default BOINC is allowed to use more memory when the machine is idle, and that should allow the tasks enough to complete. This has nothing to do with the number of days of work you would like to keep on hand.

And there is more then one way to set your preferences. Good advice is to pick one and stick with it. One way is via the project website in the participants section. The other is on the BOINC Manager. The settings in your BOINC Manager are just for the specific machine you set them on. The settings on the website will pertain to all hosts you attach to the project. In your case, where you have one host, it may be difficult to appreciate the difference. But that is likely how you set one thing and later saw another setting.

---

Rosetta Moderator: Mod.Sense

...

Now I'm thinking that the tasks are not all actually running. ...

Well, he may actually be correct and have two tasks running on the same processor.

I see that he has upgraded to BOINC 6.10.36. A few days after upgrading this happened on my two-core Intel Mac:

I left it running peacefully by itself in the afternoon. The morning sfter I found it running two Rosetta tasks on one processor (40-something each), the Finder occupying the other processor 100% (doing whatever), and the Kernel_Task process - as usual in various conflicts - being much too active. Never seen the like before.

A cold boot cured that one. But strange things have occurred also on my PPC after upgrading to 6.10.36, so I have this version under strict observation, and 6.10.35 ready to downgrade (which fixed my PPC-problems).

By the way, in Mr. Parlette's messages I observe:

> 3/12/2010 4:08:26 PM suspend work if non-BOINC CPU load exceeds 25 %
> 3/12/2010 4:08:26 PM (to change, visit the web site of an attached project,
> 3/12/2010 4:08:26 PM or click on Preferences)

He is aware of this new forced default in (at least on Mac) 6.10.35- BOINC versions, and really wants BOINC to function in this way?

I'm not disputing your identification of the version of rosetta I'm running, but how do I identify it myself? When I look at the messages, what I see is:

3/16/2010 7:31:52 PM rosetta@home Resuming task abinitio_withrelax_nodisulf_nohomfrag_nonative_cst0.1_129_B_1wgbA_SAVE_ALL_OUT_17951_1166_0 using minirosetta version 205

which doesn't exactly correspond to what you're talking about nor to what I see in the Download, install, and run BOINC accessed from the first page of the website. (The page agrees with you, of course.)

In any case, how do I downgrade to 6.10.18? I d/l it, but should I stop BOINC altogether before installing it? I'm assuming the disk status from a terminated 6.10.36 is able to be restarted with a freshly installed 6.10.18?

To further comment on the previous discussion in this thread, I know I have only 1 core/cpu. Eventually, all 3 rosetta WUs were running, albeit at alternating times, sometimes called multi-tasking, by which I mean that each independently progresses toward completion. (I have been taught that it may only be called multi-processing if you have more that one processor.) My other project, SETI, initiates one WU and runs it to completion (interspersed w/ rosetta, of course) before initiating the next SETU WU. This is the behavior I was expecting. This is the behavior I did not see w/ rosetta. In fact, now two of the rosetta WUs completed, were u/l, and reported, and the third is now progressing toward completion (83%, ~4 hrs to go).

BTW, I have yet to get any more rosetta WUs. I have 6 SETI WUs waiting in the wings. Admittedly rosetta WUs are MUCH bigger, but still . . . .

(SETI takes about 4 hrs and rosetta takes about 24 hrs for most WUs.)

Finally, do you think that 25% processor is too small for the CPU load cut-off?

---

Ross, most of Rosetta's work is sent under version 2.05, but the BOINC client that Rosetta runs within has it's own version numbers.

You can use this link to download other BOINC versions

---

Rosetta Moderator: Mod.Sense

I have upgraded to BOINC 6.10.43 and I also note that Rosetta tasks are running with minirosetta 210.

4/26/2010 4:28:34 PM rosetta@home Resuming task lrm_jorj_combined_torsion_it04_run01_A_rlbn_1jfv_SAVE_ALL_OUT_IGNORE_THE_REST_NATIVE_NOCON_19702_13_0 using minirosetta version 210

I have noted that once again multiple Rosetta WUs are advancing toward completion together. (As I have and claim only one processor, they take turns with SETI as well as each other.) This became a problem with two tasks having a near Report deadline. A Rosetta task with a more distant deadline was stealing compute time until I suspended it in favor of the tasks with the nearer Report deadline

I initially logged onto here to see how I might upgrade to minirosetta 211, but I think that would be pointless, as I expect BOINC does the assigning of tasks, not minirosetta.

This being Windows, I have the opportunity, nay the obligation, to reboot upon the changes of the wind direction, so my message queue is meager.

Perhaps this is a question which should be broached in one of the BOINC fora?

---

Yes, if you have only one CPU, and two Rosetta tasks already in progress, then any switch from SETI back to Rosetta should be allocated to the one with the first deadline. If that is not what occurs, that would be an issue for the BOINC boards.

---

Rosetta Moderator: Mod.Sense

I have upgraded to BOINC 6.10.43 and I also note that Rosetta tasks are running with minirosetta 210.

I have noted that once again multiple Rosetta WUs are advancing toward completion together. (As I have and claim only one processor, they take turns with SETI as well as each other.) This became a problem with two tasks having a near Report deadline. A Rosetta task with a more distant deadline was stealing compute time until I suspended it in favor of the tasks with the nearer Report deadline

Perhaps this is a question which should be broached in one of the BOINC fora?

This is a normal Boinc thing as long as none of the units are getting close to expiring, as units get closer to expiring then you will see them switch to the unit with the closest expiration date to crunch. As long as there is no danger of anything expiring on you, Boinc kind of does its own thing. If you run multiple projects, as you do, it will still switch between them every 60 minutes, per the default settings, and you may even see some units crunching at 'high priority' if Boinc thinks they are too close to expiring.

Yes, BOINC upgraded minirosetta to 211 all by itself. (happy-face)

The problem I have had (several times) is that when 2 or so WU are getting close to the due date (I'll call them urgent), BOINC schedules a non-urgent WU (also for Rosetta) along with them, so that a non-urgent WU is accumulating compute time along with the two urgent ones. I have now posted this on the BOINC board, although it's not nearly as active as Rosetta or SETI. I guess I'll just have to wait.

My action is to suspend the non-urgent WU when it's not in progress, along with any fresh WU (not yet started) so that all of the Rosetta compute time goes to urgent WUs. (Not wuss, that's something completely different.)

I think that I may be being (automatically) penalized because I over-estimated the time the urgent WU would take to complete and it finished before I could get back (in the morning) to resume (un-suspend) the non-urgent WU and allow it to continue. Effectively Rosetta was left w/o any work to do; all WUs were suspended.

---

Yes, BOINC upgraded minirosetta to 211 all by itself. (happy-face)

The problem I have had (several times) is that when 2 or so WU are getting close to the due date (I'll call them urgent), BOINC schedules a non-urgent WU (also for Rosetta) along with them, so that a non-urgent WU is accumulating compute time along with the two urgent ones. I have now posted this on the BOINC board, although it's not nearly as active as Rosetta or SETI. I guess I'll just have to wait.

My action is to suspend the non-urgent WU when it's not in progress, along with any fresh WU (not yet started) so that all of the Rosetta compute time goes to urgent WUs. (Not wuss, that's something completely different.)

I think that I may be being (automatically) penalized because I over-estimated the time the urgent WU would take to complete and it finished before I could get back (in the morning) to resume (un-suspend) the non-urgent WU and allow it to continue. Effectively Rosetta was left w/o any work to do; all WUs were suspended.

Next time just leave Boinc alone and see if it works it out all by itself, it should, yes you could end up with several units 'in progress' while it figures out which unit needs to run NOW, but it is designed to do that all by itself. The other thing could be your cache could be too big by just a little.

While it is possible that there is a bug in the BOINC scheduler you still haven't ruled out memory as the controlling factor.

While the amount of memory you allow BOINC to use would seem to be sufficient (1247.24MB when active) it may be that there isn't that much memory actually available at all times. And if that is the case I think you would see exactly the behaviour you have described. Mod.sense and I both posted about this earlier in the thread but what appears clear at one end may appear as gobbledegook at the other end so please ask for clarification if you need it. There are ways to see exactly how much memory is being used by which programs at a given point in time. I don't run Windows but if you ask there are several posters here who could advise you on how to see this information for your machine.

As all the WUs are completing successfully it seems the most serious concern is the potential for missing deadlines. I think I have a pretty simple solution for that: reduce your target CPU run time. The deadlines remain the same, so; less CPU time allowed within the same wall clock time allowed (time between assignment of WU to your machine to the deadline) will equal less deadline pressure.

A glance at your completed WUs suggests you currently have this value set to 1 day. In case you have forgotten where to find this go to your account page on the website and click on the rosetta@home preferences link. I wouldn't recommend reducing it to below say 10 hours in one go as this might confuse BOINC. Mind you I haven't thoroughly thought this through, this being the exact number you should be using for this setting and best practices to get there from where you are now. I'm just being cautious (maybe overly so) when I suggest not making a sudden drastic change. Or you could just make the change (I think a 8 or 10 hour setting would alleviate your deadline worries) while resolving not to make any other changes or do any sort of fiddling (like suspending WUs or setting NNT) for a couple of weeks while BOINC adjusts to the new setting.

Best,
Snags

editing just to emphasize that the change to target CPU time that I propose is not likely to change the primary behavior you have described. You will still see multiple rosetta wus in various stages of completion but you will be reducing the risk that those wus will miss their deadlines.

Snagletooth,

I did an instantaneous sample just now and minirosetta 211 was using 270,872 as well as some other similar sizes from time to time (Windows Task Manager).

I guess I didn't record enough detail, but IIRC, I had 4 rosetta WU, of which 2 were due soon (the next day or one more) the other two were due several days further down the pike. The two near WU were both coming along and one of the far WU started accumulating time as well. I waited until the less urgent WU was not running and I suspended it as well as the (also less urgent) fresh WU. Unfortunately, when I got up the next morning, the two urgent WU had completed and Rosetta was out of WU. I unsuspended the in-progress WU but I think the damage had been done. Since then I've only been getting one at a time, as earlier ones finish.

Since Rosetta doesn't have the regular down-time of SETI, this isn't so bad, but I'm definitely not getting 4 days ahead.

I do see High Priority on tasks from time to time, including the set discussed above, but it wasn't consistent, so I didn't pay as much attention to it. Specifically, sometimes an urgent WU would be High Priority and sometimes not. I was beginning to suspect a stochastic scheduling scenario.

I never saw a Waiting For Memory message, although on rare occasions I'll get a Low Virtual Memory warning (from the OS). At that point I shut everything down and pretend the wind blew, so I re-boot. I'll admit that it's not clear to me how having too little memory (virtual or not) would cause an additional WU to be started before others are completed. It's certainly true that a Rosetta WU is MUCH bigger than a SETI WU and could require more actual and virtual memory.

My lame GPU is not available for computation.

I'm not sure what you mean by "target CPU run time." Is that Computing preferences | Network usage | Computer is connected to the internet about every [ ] days? I have left it blank since I have DSL and am always connected. My most recent completed Rosetta WU took 30 hours to complete, which when shared with SETI is 60 hours or almost 3 days, since I don't use 4-6 AM for BOINC. But 3 days elapsed would seem to mitigate against claiming 1 day turn around, if that's what you're saying.

OTOH, I am no longer suspending work when the computer is in use.

---

Target runtime per task is a setting that is specific to Rosetta, so you will not find it configurable within the BOINC client. It is set via the website using the participants link at the top of this forum page. A longer preference allows more models to be generated before reporting the results.

Tip: make changes to this setting gradually over the course of a week or so, otherwise the client will not have time to adapt to the change in how long tasks take to run and can often request dramatically too much or too little work, especially if you keep a number of days of work on hand.

---

Rosetta Moderator: Mod.Sense

Target runtime per task is a setting that is specific to Rosetta, so you will not find it configurable within the BOINC client. It is set via the website using the participants link at the top of this forum page. A longer preference allows more models to be generated before reporting the results.

Tip: make changes to this setting gradually over the course of a week or so, otherwise the client will not have time to adapt to the change in how long tasks take to run and can often request dramatically too much or too little work, especially if you keep a number of days of work on hand.

---

Rosetta Moderator: Mod.Sense