Workunit suggestion (growth of the farm)

Message boards : Number crunching : Workunit suggestion (growth of the farm)
Message board moderation

To post messages, you must log in.

AuthorMessage
James

Send message
Joined: 27 Dec 11
Posts: 3
Credit: 0
RAC: 0
Message 11436 - Posted: 11 Aug 2012, 17:59:26 UTC
Last modified: 11 Aug 2012, 18:00:03 UTC

I have been attempting to get a few workunits to do some work over the past few days and all of the current rendering sessions have a RAM requirement of approx. 6GB per workunit. This seriously limits the amount of people who can work on the project (cross out ALL 32-BIT machines and the large majority of 64-bit OEM machines/laptops).

Because of this, not a lot of work gets done (and at a slow pace). Thus, my suggestion is to split work in to two different blocks - small and large. A simple classification of requiring more than 2GB of RAM can be the floor for the long workunits and anything requiring less goes in to the short queue.

People could have the chance to set which queue they pull workunits from, or have the client automatically detect the amount of RAM installed and have that determine the queue that work is pulled from.

This benefits the system three-fold:

1) More people will be enticed to join the project because people will have the specifications to do some, if not all work.

2) The queue of lined up projects will reduce faster, meaning faster turnaround times for the animators and thus a higher amount of returning "customers", meaning more work for the project on a regular basis that everyone can do.

3) Lastly, more animators will be enticed to send their projects to the server, seeing the reliable community of renderers that can complete work quickly - saving the animators money, spreading the word about the project and giving people an opportunity to donate their time to render, say, a movie/video/basic animation, etc.[/b]
ID: 11436 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
noderaser
Project donor
Avatar

Send message
Joined: 28 Mar 06
Posts: 516
Credit: 1,567,702
RAC: 0
Message 11437 - Posted: 12 Aug 2012, 7:01:10 UTC

Currently, the 6 GB sessions are the only ones actually rendering... If you're looking at the "Serverstatus" page, the sessions with lower memory requirements are awaiting admin approval and aren't rendering yet. If these sessions were rendering, hosts with less than 6 GB of RAM would be getting work from those sessions.

I can't speak to why those sessions are the only ones being rendered and others are not. I'm sure Janus has a plan for the Sunflower sessions, and there have been some user sessions submitted alongside them. However, BURP is not his full-time job and he's been known to be very busy at times.
Click here to see My Detailed BOINC Stats
ID: 11437 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
James

Send message
Joined: 27 Dec 11
Posts: 3
Credit: 0
RAC: 0
Message 11439 - Posted: 12 Aug 2012, 9:45:55 UTC - in response to Message 11437.  

Yes, I did look at the current sessions before writing the above post - I saw the smaller sessions available and thought, perhaps there should be two queues.

I have a feeling that the long workunits are to stress the server, perhaps?
ID: 11439 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile DoctorNow
Project donor
Avatar

Send message
Joined: 11 Apr 05
Posts: 403
Credit: 2,189,214
RAC: 7
Message 11440 - Posted: 12 Aug 2012, 14:06:56 UTC - in response to Message 11439.  
Last modified: 12 Aug 2012, 14:10:11 UTC

I have a feeling that the long workunits are to stress the server, perhaps?

No, not really. The server would be under a stress test when there would be big sessions with small WUs and an upload-size like the Sunflower-ones. ;-)

There is no real need for making two queues here. The server always gives out WUs which are appropriate for your system - at least should. ;-)
Life is Science, and Science rules. To the universe and beyond
Proud member of BOINC@Heidelberg
My BOINC-Stats
ID: 11440 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
James

Send message
Joined: 27 Dec 11
Posts: 3
Credit: 0
RAC: 0
Message 11442 - Posted: 12 Aug 2012, 16:25:07 UTC - in response to Message 11440.  

I have a Linux system with enough RAM to run the smaller workunits currently in the queue, HOWEVER there is no Linux app just for the Blender renderer.

This means currently the project will not allow me to perform any work.

Janus, could we please get a regular Blender app for Linux? I wish to do some work but I can't.
ID: 11442 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile DoctorNow
Project donor
Avatar

Send message
Joined: 11 Apr 05
Posts: 403
Credit: 2,189,214
RAC: 7
Message 11443 - Posted: 12 Aug 2012, 16:57:25 UTC - in response to Message 11442.  
Last modified: 12 Aug 2012, 16:58:38 UTC

HOWEVER there is no Linux app just for the Blender renderer.

That can't be, I crunched WUs for the regular Blender-app a (long) while ago under Linux (currently only crunching Windows). :-D
And looking on the apps- and download-urls I see two normal blender-apps for 32- and 64-bit, so you should have no worries.
Maybe the server gets a little problem nevertheless to distribute correct wus to you...
Life is Science, and Science rules. To the universe and beyond
Proud member of BOINC@Heidelberg
My BOINC-Stats
ID: 11443 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
noderaser
Project donor
Avatar

Send message
Joined: 28 Mar 06
Posts: 516
Credit: 1,567,702
RAC: 0
Message 11446 - Posted: 14 Aug 2012, 3:56:33 UTC

The apps are distributed automatically through the BOINC client; if you're not getting any work, it could just be that there isn't any to send out at the time. Even though a session may have a low completion percentage, there won't necessarily be work to send out due to the number of tasks created in the initial replication. As those reach their deadlines or come back unsuccessful, "repair" units are sent out to augment them. With some of these "slow" sessions it may take a while for all of the repair units to get sent out. This is done over a brute-force method (continuing to send out WUs after the initial replication) to prevent wasted resources.

There have also been some issues with certain distributions of Linux not having the necessary libraries, although IIRC if this is the case you will get work but the units will fail with an error making reference to some missing files.
Click here to see My Detailed BOINC Stats
ID: 11446 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Number crunching : Workunit suggestion (growth of the farm)