But the Host and User equivalents are getting smaller and smaller - something's taking a salami slicer to them! I'll let it run a couple more days, then I'll bury my head in that bloody C++ script again.

Are the Host & User values a weighted average over time?

Are the Host & User values a weighted average over time?

No.

BOINC tracks two values: total and RAC, for any of these groupings. RAC (aka 'average' and 'weight' internally) is the decaying one, with a half-life of seven days (but missing for Hosts at the moment): Total (which is the one I'm tracking here) should be monotonically increasing since the dawn of time. No decay.

Looking at your table from the previous message, the Host & User numbers are consistent with each other. If I add up the User increments I get 2,277 which matches the Host increment. So the issue is the difference between the Task values & the (Host/User) values. Maybe the Task value is the 'raw' number before some kind of scaling/adjustment is being applied? (you probably know this, I'm catching up :)

As you said, would need to look in the code to see what's happening. This also assumes these number are being computed at the same time and it's not due to two different credit scripts running at different times?

OK, here's today's show.

Task 22347460													

Trickle	Date		Host total	Host increment		User total	incr.	Delta		Task total	Task increment	Interval

			93,562					33,375,985							
1	15/10/2023						33,376,834	849			  849.83	849.83		
2	15/10/2023						33,377,585	751	98		1,678.16	828.33		09:17
3	16/10/2023	95,839		2,277			33,378,262	677	74		2,506.49	828.33		09:19
4	16/10/2023	96,441		  602			33,378,864	602	75		3,334.82	828.33		09:07
5	17/10/2023	96,969		  528			33,379,391	527	75		4,163.15	828.33		09:06
6	17/10/2023	97,422		  453			33,379,845	454	73		4,991.48	828.33		09:23
7	17/10/2023	97,800		  378			33,380,223	378	76		5,819.81	828.33		09:32
8	18/10/2023	98,104		  304			33,380,527	304	74		6,648.14	828.33		09:12

I've added two extra columns - Delta, to get an idea of the rate of change, and interval (hh:mm), the time between trickles. All the figures have been copied directly from this web-site, and thus from the project database. The 'Date' is actually the full timestamp from the 'Trickle received' column in the task display: I've stopped this machine from making any extra scheduler contacts, so the trickles are sent and received within seconds of creation.

Like Glenn, I first assumed that the decay function for RAC had somehow been drawn into the calculation of totals. But that would be an exponential decay, and this is looking more like a straight line. There's the slightest hint that delta is related to interval, but it's not strong enough to be certain.

Time for that code-walk. First, I have to find that download link again - make sure no-one's been fiddling with the script while we haven't been watching. Edit - found it. No change except the rather feeble comment on the correction factor.

I won't repeat the whole table now, but it's just added a new line:

12	19/10/2023		98,575	5		33,380,998	6	74		9,961.47	828.33		09:13

Which begs the question - what happens when the increment reaches zero? Does it stop, or go negative? The answer will happen around 02:40 UTC in the morning*, but I won't be around to see it. Anyone want to start a sweepstake on the answer?

* Or it may not happen at all. I'm just entering a severe weather warning, with thunderstorms, which sometimes trigger blackouts - and that host isn't protected by a UPS.

Well, we were spared the lightning, so the power supply and the task have survived so far. But the wind and rain were horrendous - everything which can leak, is leaking.

But back on topic ...

Task 22347460													

Trickle	Date		Host total	Host increment		User total	incr.	Delta		Task total	Task increment	Interval

			93,562					33,375,985							
1	15/10/2023						33,376,834	849			  849.83	849.83		
2	15/10/2023						33,377,585	751	98		1,678.16	828.33		09:17
3	16/10/2023	95,839		2,277			33,378,262	677	74		2,506.49	828.33		09:19
4	16/10/2023	96,441		602			33,378,864	602	75		3,334.82	828.33		09:07
5	17/10/2023	96,969		528			33,379,391	527	75		4,163.15	828.33		09:06
6	17/10/2023	97,422		453			33,379,845	454	73		4,991.48	828.33		09:23
7	17/10/2023	97,800		378			33,380,223	378	76		5,819.81	828.33		09:32
8	18/10/2023	98,104		304			33,380,527	304	74		6,648.14	828.33		09:12
9	18/10/2023	98,334		230			33,380,757	230	74		7,476.48	828.34		09:15
10	18/10/2023	98,489		155			33,380,912	155	75		8,304.81	828.33		09:04
11	19/10/2023	98,570		81			33,380,992	80	75		9,133.14	828.33		09:03
12	19/10/2023	98,575		5			33,380,998	6	74		9,961.47	828.33		09:13
13	20/10/2023	98,643		68			33,381,065	67	-61		10,789.80	828.33		09:09

Well - I never anticipated that - it's bounced!

That's actually quite useful, because it points the finger at a very specific line in the code, which I'd already picked out as looking dodgy. I'll pass it up the line.

I see now some accumulated RAC on my Hosts page. Numbers are small and do not correspond to the user RAC. But maybe they'll catch up.

---

I see now some accumulated RAC on my Hosts page. Numbers are small and do not correspond to the user RAC. But maybe they'll catch up.

Andy was working on the credit script yesterday, and deployed a modification to address the RAC part of the problem. But that depends (properly) on the amount of "new work", which is the column I've labelled 'increment' in my tables.

For trickle 12 yesterday, with increment 5, that earned me a RAC of 0.03: with trickle 13, increment 68, RAC's gone up to 6.60. But it should be based on the full 828.33 awarded for the trickle.

Yes, I see that I now have a +ve average credit on one of my BOINC instances. I won't attempt to confuse the CPDN BOINC server by letting my Wine instance upoload zips and trickles for the same tasks to satisfy my curiosity about how it would handle same!

OK, the task under the microscope has completed and reported success. Here are the figures for the final trickle: there was no further change on report.

The totals were:

Task: 19,901.44 credits
Host and User: 10,744 credits added.

There's something very, very wrong there. This is how it happened:


The Task line (yellow) is exactly correct, according to theory.
The Host line (exactly matched by the User line - pink) is the one we need to look into more deeply. It even gave me too much for the final trickle, but that doesn't make up for the dip in the middle.

It was very helpful that RAC was fixed just at the right time: that line looks pretty good for RAC based on the host increments.