Rosetta - Resource Hog?

Anything "BOINC" specific can be commented on here...such as Project news and announcements etc. Also: any problems with BOINC or maybe you have found something interesting, tell us about it. Chat about the various 3rd party client applications used for some of the projects such as optimised clients.
Post Reply
Torn Jacketson
Posts: 28
Joined: Sun Jan 31, 2016 7:05 pm

Rosetta - Resource Hog?

Post by Torn Jacketson »

Hi guys,

I have started running a few other projects now, all the ones indicated by UBT - Timbo that are non CPU intensive, and also Einstein and Rosetta.

Since adding Rosetta I have noticed that it is a bit of a resource and BOINC hog. What I mean is that it has killed all the other projects stone-dead. Their tasks never seem to get to run - Rosetta is hogging all 12 processors all the time. It is only ever Rosetta tasks that are running on my system now. SETI and friends never get to play with the ball!

I have ensured that the "Resource Share" is configured appropriately - no change. Rosetta still does not play fair. I have even reduced Rosetta down to 1/10th of all the others - still it hogs. The only way I can stop it is to suspend the project, but as soon as I Resume and a task slot becomes available it is Rosetta that leaps in and soon it occupying all 12 slots again.

Fair dues, it is racking up credits quickly, but to the detriment of all the others. The only credits I am now getting from the other projects is from the "Pending" or "Inconclusive" streams.

Anyone else noticed this - or have any idea why Rosetta is behaving this way on my machine ?

Steve (aka Torn Jacketson)
Torn Jacketson
Posts: 28
Joined: Sun Jan 31, 2016 7:05 pm

Re: Rosetta - Resource Hog?

Post by Torn Jacketson »

... in fact, did not realise this before, when I suspend Rosetta SETI and friends get to run. As soon as I resume Rosetta the tasks for the other projects are immediately put into "Waiting to run" and the Rosetta tasks start running again. Rosetta is definitely not playing fair and if I cannot find a reason for this it is gonna get canned

Steve
Torn Jacketson
Posts: 28
Joined: Sun Jan 31, 2016 7:05 pm

Re: Rosetta - Resource Hog?

Post by Torn Jacketson »

One thought - does BOINC schedule base on the WU deadline? This may be why it is happening. Rosetta is issuing me WU's with a deadlines of a few days, whilst the others are issuing WU's with deadlines a few weeks away ... In this case the "Resource Share" is not really a proper share at all - projects can bypass that by issuing WU's with short deadlines as they will always get preferential scheduling - which is what Rosetta seems to be doing .... IF that is the way BOINC schedules!

Steve
UBT - Mikee
Marvin the Dalek
Posts: 4395
Joined: Wed Mar 15, 2006 12:00 am
Location: North Wales

Re: Rosetta - Resource Hog?

Post by UBT - Mikee »

Hi

Seems you've spotted the answer! BOINC Manager does indeed schedule WUs with the shortest deadline. As far as I'm aware the only way round is to suspend but they'll need to finish before the dealine, else they'll get reissued. If you want to the only way round is to finish the WUs you have and set to no new tasks, then when you want to do a batch more you can get more work. Try to ensure that the cache is set relatively low so you only get a few Rosetta WUs at a time.

Mike
Follow us on Twitter... http://twitter.com/UKBOINCTeam

Image Image

Image
Torn Jacketson
Posts: 28
Joined: Sun Jan 31, 2016 7:05 pm

Re: Rosetta - Resource Hog?

Post by Torn Jacketson »

Thanks for the confirmation Mike. I came to the conclusion it had to be that, and I have done exactly what you suggest, I will let Rosetta run its tasks through without getting new ones and the other projects will just have to take the hit.

Steve
Woodles
UBT Contributor
Posts: 11757
Joined: Thu Dec 20, 2007 12:00 am
Location: Cambridgeshire

Re: Rosetta - Resource Hog?

Post by Woodles »

Resource share no longer decides the amount of resource used. It now attempts to balance the total credit earned over a time period so if you add a new project with a share of 1 to your long running project with a share of 100, the new project will get all the work until it's caught up.

As you've noticed and Mike has confirmed, it also schedules depending on deadline. Not as straightforward or, in my opinion, as useful as before.

An alternative to manually stopping and starting projects is to limit Rosetta to a maximum number of cores and leave the rest for the other projects. In this way, you can set up your own, crude resource sharing.

Create an app_config.xml file in the boinc.bakerlab.org_rosetta project directory (or add to the existing one if you're already using an app_config.xml) consisting of:

Code: Select all

<app_config>
 <project_max_concurrent>N</project_max_concurrent>
</app_config>
Where N is the maximum number of cores you want Rosetta to use.

If you're using a Boinc version earlier than 7.4.9 then you'll need to add a limit for each application individually instead (found in the client_state.xml file - if I remember right, Rosetta only has minirosetta applications)

Code: Select all

<app_config>
 <app>
  <name>minirosetta</name>
  <max_concurrent>N</max_concurrent>
 </app>
</app_config>
Again, N is the maximum number of cores that the application is allowed to use.

The drawback being that even if your other projects run out of work and there's no new ones downloaded, Rosetta won't pick up the spare cores.

Regards,

Mark
Torn Jacketson
Posts: 28
Joined: Sun Jan 31, 2016 7:05 pm

Re: Rosetta - Resource Hog?

Post by Torn Jacketson »

Thanks for that Mark - I will have a play with the config file. I need to reign in Rosetta as it has just killed any progress with SETI. I will let the current batch of Rosettas run through so that I meet the deadlines and then control it a bit. Just think it is a bit "nasty" that a project can keep issuing WU's with tight deadlines and effectively rob all of the computing resource away from other projects .... just seems a little underhand somehow.

Steve
Woodles
UBT Contributor
Posts: 11757
Joined: Thu Dec 20, 2007 12:00 am
Location: Cambridgeshire

Re: Rosetta - Resource Hog?

Post by Woodles »

Hi Steve,

No problem. I had the same issues with SRBase repeatedly issuing work with a 24 hour deadline, nothing else got crunched until it became critical and then the recently downloaded SRBase WUs were also critical so it was hit and miss what actually got finished before the deadlines. I now have SRBase limited to a maximum of 2 out of 4 cores and everything seems to be playing nicely so far.

Forgot to mention that limiting particular applications (not project) to a number of cores could lead to more cores used than you think if two applications from the same project are running at the same time. SrBase has applications called srbase, srbase1, srbase2 etc and limiting them to 2 cores leads to occasionally 6 cores being used, 2 for srbase, 2 for srbase1 and 2 for srbase2. Sadly, my Linux version of Boinc hasn't got the capability of limiting the project as a whole.

Mark
UBT - Timbo
UBT Forum Admin
Posts: 9673
Joined: Mon Mar 13, 2006 12:00 am
Location: NW Midlands
Contact:

Re: Rosetta - Resource Hog?

Post by UBT - Timbo »

Torn Jacketson wrote:One thought - does BOINC schedule base on the WU deadline? This may be why it is happening. Rosetta is issuing me WU's with a deadlines of a few days, whilst the others are issuing WU's with deadlines a few weeks away ... In this case the "Resource Share" is not really a proper share at all - projects can bypass that by issuing WU's with short deadlines as they will always get preferential scheduling - which is what Rosetta seems to be doing .... IF that is the way BOINC schedules!

Steve
Hi Steve,

It looks like you are well on your way to understanding how BOINC Manager works and what the "resource share" function is all about.

In my case, RS hardly works at all....so, I manually adjust some projects to "no new work" in order to allow them to run down what work they have and then I can "allow new work" when required. It also help to set the "Minimum work buffer" to quite a low number - I usually leave it around 0.75...but even so, some of the shorter running tasks still fill up the cache.

Recently, I was running some SETI CPU tasks and even though they have a long deadline, they were still running despite me having lots of other tasks with shorter deadlines :( You'd have thought the earlier deadline tasks would have completed sooner ;)

To this end, I use BOINCTasks*, which has a few useful "extra" functions over BM, to allow you to "control" what's going on. One really useful function is that it groups all of the same tasks (from any specific project) all together. So, if I look at what is going on now, it lists that I have:

Code: Select all

PrimeGrid	1.39 PPS (Sieve) (cudaPPSsieve)	80 [Tasks], double click to expand 	- (-)	0.00	0.00	23:13:20	04d,19:33:02	0.0424C + 1NV	Ready to start	0.00	UBT - Timbo
So, I can see I have 80 (GPU) tasks for PrimeGrid, which total 23:13:20 (h/m/s) of "work" to be done, and the deadline is in 4days 19 hrs...it also shows I need 0.0424 of my CPU to run these, plus 1x NVidia core.

The useful bit is if I want to suspend ALL of these, I just right click on this line and the suspend option is there...so no need to "order" all the tasks in BM, and then click the first and the last and then hit suspend.

Likewise, it's easy to enable or disable any of the other tasks just as easily....either on a "task basis" or on an entire "project basis".


As far as Rosetta goes, I've never found it to use up too many resources and I've never found it has interfered with WUProp or Goofy's...however, I always leave at least 1x CPU "free" so that there are resources available for GPU apps and the NCI's.

So, if you have a 12-core CPU, perhaps try setting the "On multiprocessor systems, use at most" to no more than "92" (i.e. 100% - 1/12th, which is 8.33) which will allow BOINC to use at most 11 of the 12 CPU's. This might also allow the NCI tasks to crunch concurrently, as they do need some CPU power but not much.

regards
Tim

* = BOINC Tasks can be downloaded from here: http://efmer.com/b/boinctasks
UBT - Timbo
UBT Forum Admin
Posts: 9673
Joined: Mon Mar 13, 2006 12:00 am
Location: NW Midlands
Contact:

Re: Rosetta - Resource Hog?

Post by UBT - Timbo »

Torn Jacketson wrote:... in fact, did not realise this before, when I suspend Rosetta SETI and friends get to run. As soon as I resume Rosetta the tasks for the other projects are immediately put into "Waiting to run" and the Rosetta tasks start running again. Rosetta is definitely not playing fair and if I cannot find a reason for this it is gonna get canned

Steve
Hi Steve

As a further update I found this message on the Rosetta forum:

https://boinc.bakerlab.org/rosetta/foru ... true#80045

So, it seems that Rosetta does rather like a lot of spare "real" CPU's all to itself (as opposed to HT CPU's)...so, it might be an idea to lower the number of CPU's in use running only Rosetta and see if that improves your throughput...

regards
Tim
Torn Jacketson
Posts: 28
Joined: Sun Jan 31, 2016 7:05 pm

Re: Rosetta - Resource Hog?

Post by Torn Jacketson »

Thanks Chaps.

Rosetta was being a real hog, so eventually I aborted its remaining tasks. I now need to fully understand what I need to tweak to control that beast. For now I have set it to no new work. It can stay on the naughty step for a while.

Steve
Torn Jacketson
Posts: 28
Joined: Sun Jan 31, 2016 7:05 pm

Re: Rosetta - Resource Hog?

Post by Torn Jacketson »

Mark,

I have created an app_config.xml as you suggested (location is D:\BOINC\DATA\projects\boinc.bakerlab.org_rosetta) and the only contents are as follows:

Code: Select all

<app_config>
 <project_max_concurrent>1</project_max_concurrent>
</app_config>
I tried to reel in Rosetta as much as I could to see if this worked, with the above, but as soon as I released the Kraken it swamped my machine again and took over all 12 active process slots. I aborted them all and then restarted my machine to ensure that BOINC was restarted clean with the XML file in place, but the same happened - Rosetta just downloaded a load of WU's and took over all slots.

Have I missed something or typed something wrong as the setting has seemingly been totally ignored?

I am running BOINC 7.6.22 so did not bother with the extra lines you suggested as I am running a version after the one you stated.

Hope you can point out where I went wrong otherwise Rosetta is for the permanent chop as it does not play fair.

Steve
Woodles
UBT Contributor
Posts: 11757
Joined: Thu Dec 20, 2007 12:00 am
Location: Cambridgeshire

Re: Rosetta - Resource Hog?

Post by Woodles »

Hi Steve,

Assuming that's your current BOINC data directory then that's the right location and all the contents you need.

I've just cut and pasted your lines into a blank file in SRBase and it worked as expected, the project went from eight running tasks to one with seven waiting to run.

A thought, once you've created the file you need to get BOINC to read it. Did you stop and restart BOINC manager or use the "Read config files" option on the "Tools" menu? If you have a look at the BOINC log ("Tools", "Event log") then you should see an entry along the lines of:

Code: Select all

31/05/2016 08:22:06 |  | Re-reading cc_config.xml
31/05/2016 08:22:06 |  | cc_config.xml not found - using defaults
31/05/2016 08:22:06 |  | log flags: file_xfer, sched_ops, task
31/05/2016 08:22:06 | SRBase | Found app_config.xml
31/05/2016 08:22:10 | SRBase | Message from task: 0
Obviously, it's just the

Code: Select all

31/05/2016 08:22:06 | SRBase | Found app_config.xml
bit that you're looking for :)

[EDIT]
Just finished adding Rosetta to this machine (some very big files to download there!) and it can be restricted to however many cores I specify.
[/EDIT}

Regards,

Mark
Torn Jacketson
Posts: 28
Joined: Sun Jan 31, 2016 7:05 pm

Re: Rosetta - Resource Hog?

Post by Torn Jacketson »

Hi Mark,

Glad that I did nothing wrong. I rebooted my machine so I thought that would ensure a clean restart of BOINC, but maybe not. When I am home tonight I will check the logs and use the Read Config files as you suggest.

Steve
Woodles
UBT Contributor
Posts: 11757
Joined: Thu Dec 20, 2007 12:00 am
Location: Cambridgeshire

Re: Rosetta - Resource Hog?

Post by Woodles »

Hi Steve,

Nope, it all looks right to me, I'm surprised that a reboot wouldn't re-read the files.

I tend to 'personalise' my BOINC installations so the data directory might not be the right one for you, it should be the one with the "minirosetta_3.73_windows_x86_64.exe" file already in it ... assuming you're running a 64 bit version of Windows :)

Once it's in the right place, there should be an entry in the log as above showing that the app_config.xml file has been found.

Regards,

Mark
Torn Jacketson
Posts: 28
Joined: Sun Jan 31, 2016 7:05 pm

Re: Rosetta - Resource Hog?

Post by Torn Jacketson »

Hi Mark,

I have taken the Read Config Files option and the Event log shows that it has read the config file from the directory now, so hopefully that is all in place.

I will release Rosetta soon and see what happens!

Thanks,
Steve
Woodles
UBT Contributor
Posts: 11757
Joined: Thu Dec 20, 2007 12:00 am
Location: Cambridgeshire

Re: Rosetta - Resource Hog?

Post by Woodles »

Hi Steve,

Good luck, if it's found the file then you're most of the way there.

Regards,

Mark
Torn Jacketson
Posts: 28
Joined: Sun Jan 31, 2016 7:05 pm

Re: Rosetta - Resource Hog?

Post by Torn Jacketson »

Hi Mark,

Rosetta is behaving now!! Asteroids went a bit wild when the server finally released me some WU's so I have done the same to that.

Thanks,
Steve
Woodles
UBT Contributor
Posts: 11757
Joined: Thu Dec 20, 2007 12:00 am
Location: Cambridgeshire

Re: Rosetta - Resource Hog?

Post by Woodles »

Hi Steve,

Glad to hear it's all under control now.

The only thing to watch out for is running out of work on your other projects as Rosetta won't pick up the slack any more. Other than that, happy crunching :)

Regards,

Mark
Torn Jacketson
Posts: 28
Joined: Sun Jan 31, 2016 7:05 pm

Re: Rosetta - Resource Hog?

Post by Torn Jacketson »

Fingers crossed that won't happen. I have a couple of days in reserve so that will keep me going .... unless they all go the way of the Asteroids server over the past couple of weeks I will hopefully be okay!

Steve
Post Reply