This has been reported before by various people. The Result Duration Correction Factor is going up, instead of down. And with the RDCF going up, so does the ETC for the result, which gives less results to download.

This has been happening for a couple of builds now. I'm now using 5.8.6a on Windows 2000.

I'm now timing Primegrid LLR results.
So far I have this (small) list:

Primegrid 30-01-2007
Starting ETC 8:35 - RT 5:09 (didn't record RDCF)

<duration_correction_factor>0.584925</duration_correction_factor>
ETC 8:37 - RT 5:25

<duration_correction_factor>0.586248</duration_correction_factor>
ETC 8:38 - RT 5:09

<duration_correction_factor>0.587997</duration_correction_factor>
ETC 8:40 - RT 5:13

<duration_correction_factor>0.589613</duration_correction_factor>
ETC 8:41 - RT 5:14

ETC = Estimated time to completion
RT = Real time to completion

As you see the real times to completion are way lower than the estimates. Yet the estimates go UP.
The RDCF in client_state.xml also go up.

So if nothing changed in the RDCF calculation code, then what did change that our RDCF's go up, instead of down?

(I am posting it here for now, as I can't reach the Alpha email list as Gmail is down ;))

---

The inclusion of Advanced local preference in 5.8.5 was an error. It's not ready to be tested yet. That is why it's not in 5.8.6a.

David put in a fix for the next version of BOINC.

David 30 Jan 2007
- core client: computation of duration_correction_factor was messed up.
E.g. when the correct value is .5, it converges to sqrt(.5)

---

David put in a fix for the next version of BOINC.

David 30 Jan 2007
- core client: computation of duration_correction_factor was messed up.
E.g. when the correct value is .5, it converges to sqrt(.5)

The fix does not stop the DCF from increasing rapidly if it is too low, doing so would under many circumstances make work be late. Please talk to the administrators of the project to get different estimates for the different types of tasks.

---

BOINC WIKI

David put in a fix for the next version of BOINC.

David 30 Jan 2007
- core client: computation of duration_correction_factor was messed up.
E.g. when the correct value is .5, it converges to sqrt(.5)

The fix does not stop the DCF from increasing rapidly if it is too low, doing so would under many circumstances make work be late. Please talk to the administrators of the project to get different estimates for the different types of tasks.

Which "fix" is included in the Public Release of 5.8.8, John?

---

Which "fix" is included in the Public Release of 5.8.8, John?

I suppose the last one, written and reported by David? IIRC (can't find it anymore) the John's one did not yet get into release branch.

Peter

OK, found it (1 Feb 2007 16:28:12 -0500):

It should have the DCF calcs used in 5.8.0.

----- Rom

Which "fix" is included in the Public Release of 5.8.8, John?

The correct one got it today to 5.8.9.

Peter

I've now been running 5.8.9 for a couple of hours and see the results already. Primegrid's ETC's are now around 5m 28s and still going down, instead of 8m 40s and going up. :-)

---

Ageless,
I've had the same problem on SIMAP:
Initial RDCF: 0.680, ETC: 2h20m,
RT: 1h38 for every WUs and after 15-20 WUs,
Final RDCF: 0.714, ETC: 2h24m;

As a bonus problem, I experienced this!

Bad times for periodic projects! :-(

---