BOINC is bad at math, reason why some people can't keep a full cache?

I'm using BOINC 7.0.31.

Here is a recent scheduler request:

BOINC requested around a day's worth of work, and claims to have received over 41 hours worth of work.

But it actually only got around 12 hours worth of work.


So if it is thinking that it got more than it actually did (41 hours vs 12 hours), is this why people are having problems maintaining a full cache?
____________


Donate with your searches and online buys:
http://www.goodsearch.com/toolbar/university-of-california-setihome

My first thought would be along the lines that your system is still learning how long these tasks will take. The server shows you only have 2 tasks completed for AP on the GPU.

It could also be that the server did something weird with your TDCF. I noticed that it doubled on one of my machines for several hours the other day for what seemed like no reason.

Another guess is that the task duration also includes your work fetch values.
____________
SETI@home classic workunits: 93,865 CPU time: 863,447 hours

Join the BP6/VP6 User Group today!

I still say BOINC is bad at math.

I have my cache settings at 3+.01 and I currently have about 26.5 hours of work according to the 'Remaining' column totals, yet I get this in my log:

2012-07-10 07:13:46 PM | SETI@home | Not requesting tasks: don't need

____________


Donate with your searches and online buys:
http://www.goodsearch.com/toolbar/university-of-california-setihome

I still say BOINC is bad at math.

I have my cache settings at 3+.01 and I currently have about 26.5 hours of work according to the 'Remaining' column totals, yet I get this in my log:

2012-07-10 07:13:46 PM | SETI@home | Not requesting tasks: don't need

Are those 26.5 hours of work for SETI or for all your projects?

Gruß,
Gundolf

I still say BOINC is bad at math.

I have my cache settings at 3+.01 and I currently have about 26.5 hours of work according to the 'Remaining' column totals, yet I get this in my log:

2012-07-10 07:13:46 PM | SETI@home | Not requesting tasks: don't need

One thing which often makes such difficulties is that the local preferences override the web preferences, and it's fairly easy to unintentionally have local preferences in effect. Unless you've set that 3+.01 from the BOINC Manager advanced menu | Preferences it may not actually be in effect.

Joe

I did a test on PrimeGrid with 7.0.31

7/11/2012 3:05:10 PM PrimeGrid [sched_op] Starting scheduler request
7/11/2012 3:05:10 PM PrimeGrid Sending scheduler request: Project initialization.
7/11/2012 3:05:10 PM PrimeGrid Requesting new tasks for CPU and NVIDIA
7/11/2012 3:05:10 PM PrimeGrid [sched_op] CPU work request: 1.00 seconds; 0.00 devices
7/11/2012 3:05:10 PM PrimeGrid [sched_op] NVIDIA work request: 1.00 seconds; 0.00 devices
7/11/2012 3:05:15 PM PrimeGrid Scheduler request completed: got 1 new tasks
7/11/2012 3:05:15 PM PrimeGrid [sched_op] Server version 613
7/11/2012 3:05:15 PM PrimeGrid Project requested delay of 7 seconds
7/11/2012 3:05:15 PM PrimeGrid [sched_op] estimated total CPU task duration: 0 seconds
7/11/2012 3:05:15 PM PrimeGrid [sched_op] estimated total NVIDIA task duration: 9921 seconds
7/11/2012 3:05:15 PM PrimeGrid [sched_op] Deferring communication for 7 sec
7/11/2012 3:05:15 PM PrimeGrid [sched_op] Reason: requested by project

It was a new instance of BOINC and I had just attached to PrimeGrid. Asking for 1 second of work for the initial request is normal.

If you have a project as a backup project, resource share 0, BOINC will only ask the backup project for work to keep idle devices busy. Instead of building up a cache on the other project.

The object here was to test the "estimated total NVIDIA task duration" nothing more.
____________
SETI@home classic workunits: 93,865 CPU time: 863,447 hours

Join the BP6/VP6 User Group today!

It was a new instance of BOINC and I had just attached to PrimeGrid. Asking for 1 second of work for the initial request is normal.

The object here was to test the "estimated total NVIDIA task duration" nothing more.

Actually, all these quantities reflect average availability
(on_frac*active_frac, or on_frac*gpu_active_frac in the case of GPUs),
except for "estimated total NVIDIA task duration"
which (confusingly) doesn't; I fixed this.

A request for 327854 seconds of NVIDIA jobs means we want enough jobs
so that their estimated wall time is at least 327854 seconds.
In this case the estimated wall time per job was about 105000 seconds,
so the scheduler sent 4 of them.

-- David

The object here was to test the "estimated total NVIDIA task duration" nothing more.

There was a discussion a couple of months ago on boinc_alpha which touched on this.

Actually, all these quantities reflect average availability
(on_frac*active_frac, or on_frac*gpu_active_frac in the case of GPUs),
except for "estimated total NVIDIA task duration"
which (confusingly) doesn't; I fixed this.

A request for 327854 seconds of NVIDIA jobs means we want enough jobs
so that their estimated wall time is at least 327854 seconds.
In this case the estimated wall time per job was about 105000 seconds,
so the scheduler sent 4 of them.

-- David

Note that the debug messages (including 'NVIDIA task duration' in the latest BOINC clients) have that correction to wall time, but the runtime estimates displayed by BOINC Manager don't. So, if a task actually takes one hour to run, but is only active 50% of the time, the interval between 'start' and 'finish' will be two hours - and the 'NVIDIA task duration' will need to be 7200 seconds to calculate cache size correctly.