What is Rolling Average? - Page 2

garo · 2007-09-25, 16:02

It sounds like RollingAverage is not being recalculated when the program runs for less hours than specified. That would explain why it is never revised downward except when the CPU is being shared by other processes. v25.5 is currently being alpha tested. Can you check if this behaviour also occurs with v25.5? The save files are not compatible so make sure you install v25.5 in another directory.

Brian-E · 2007-09-26, 17:31

Quote:

Originally Posted by garo

It sounds like RollingAverage is not being recalculated when the program runs for less hours than specified. That would explain why it is never revised downward except when the CPU is being shared by other processes. v25.5 is currently being alpha tested. Can you check if this behaviour also occurs with v25.5? The save files are not compatible so make sure you install v25.5 in another directory.

OK, I'll suspend the current work for a while and test v25.5 . Will certainly keep it separate from the real work as you suggest. I'll report back when it's clear if there's any difference.

Brian-E · 2007-09-29, 11:17

Tested v25.5 for a couple of days running for 8 hours 30 minutes per day while CPUHours was set at 6. The file local.txt (as it is now called instead of local.ini) was rewritten several times but RollingAverage remained unchanged at 1000. This is certainly different from the behaviour of v24.14 which would have nudged RollingAverage up at an early stage. It's probably reasonable as well for the new version not to have changed RollingAverage yet because the program has not been running long enough to make a good estimate.
Now I have decided to change behaviour and run v25.5 for only a short time each day or sometimes not at all to see if RollingAverage eventually gets adjusted below 1000. Will report results.

I have one question/worry. Naturally it was important to run v25.5 in the same way as the way I was running v24.14 so that differences in behaviour could be confidently attributed to the different version of the software. However there is one fundamental difference: I am testing "dummy" exponents put into a file worktodo.txt without any communication with the server. I deliberately decided to do that because I didn't want to disrupt any genuine allocations of exponents when I'm only testing the new alpha software. And I thought that the calculation of RollingAverage should not be affected by this difference. But is that correct? Could it be that RollingAverage has remained at 1000 because the software is set up not to communicate with primenet? Does anyone know?

Mini-Geek · 2007-09-29, 11:23

Quote:

Originally Posted by Brian-E

I have one question/worry. Naturally it was important to run v25.5 in the same way as the way I was running v24.14 so that differences in behaviour could be confidently attributed to the different version of the software. However there is one fundamental difference: I am testing "dummy" exponents put into a file worktodo.txt without any communication with the server. I deliberately decided to do that because I didn't want to disrupt any genuine allocations of exponents when I'm only testing the new alpha software. And I thought that the calculation of RollingAverage should not be affected by this difference. But is that correct? Could it be that RollingAverage has remained at 1000 because the software is set up not to communicate with primenet? Does anyone know?

I'm not sure, but that might have affected it. It helps calculate the ETA, and it might only need that when on PrimeNet. If 25.5 can be run on the v4 server, I suggest you get it on and have it request TF. When you finish with the test, you can always unreserve the exponent. If it doesn't work with the v4 server, PM Prime95 for the line to make it contact the v5 server, and request TF.

Brian-E · 2007-09-30, 09:45

Quote:

Originally Posted by Mini-Geek

I'm not sure, but that might have affected it. It helps calculate the ETA, and it might only need that when on PrimeNet. If 25.5 can be run on the v4 server, I suggest you get it on and have it request TF. When you finish with the test, you can always unreserve the exponent. If it doesn't work with the v4 server, PM Prime95 for the line to make it contact the v5 server, and request TF.

Thanks. I can see the logic now that there would be no need to update RollingAverage if it is not used to give estimated completion dates.
I'll PM Prime95 for the secret line and start the testing again properly.

garo · 2007-09-30, 23:17

No don't ask for the secret line as that would get you TFs which again might not be so good for calculating the RollingAverage. Instead, communicate normally with primenet using v25.5. Then observe what happens over a week.

Brian-E · 2007-10-01, 15:46

Quote:

Originally Posted by garo

No don't ask for the secret line as that would get you TFs which again might not be so good for calculating the RollingAverage. Instead, communicate normally with primenet using v25.5. Then observe what happens over a week.

Right, George Woltman has independently said much the same in reply to my request for the secret line. Thankyou Garo and Mini-Geek for the help. I'll test v25.5 with LL testing using exponents given by the old server and will report the behaviour of RollingAverage in due course.

Brian-E · 2007-10-09, 17:39

Either the RollingAverage is not working in v25.5 or it is waiting until more data than the one week of testing that has so far been gathered before adjusting itself. During the last 7 days I have run LL testing for about 23 hours in total with CPUHours set at 6. RollingAverage should be about 500-600 on that basis but it has remained at 1000. The server was contacted several times with estimated completion dates, all apparently based on 6 hours per day running time.
I will continue testing for a while with CPUHours=6 and real running time much less than that, and later I plan to change the test by setting CPUHours=1 and testing for significantly more than 1 hour per day.
If anyone has their own testing suggestions please let me know.
Brian.

Brian-E · 2007-10-13, 10:31

Tested a few more days with CPUHours set at 6 but running less than 1 hour per day. RollingAverage remained at 1000 and the server was often contacted with ETA dates still based on 6 hours per day. Today I decided to change over to running longer than the given CPUHours instead of shorter times but encountered the strange problem that when I tried to unreserve the exponent from the menu (I wanted to start completely as new with a new exponent so that the previous testing would hopefully not affect the new testing), after apparently succeeding the v25.5 then immediately requested new work from the server and was given... the same exponent as it had just unreserved! Further testing then continued on this exponent from where it left off!

At this point I decided there isn't much point in continuing the test anyway because RollingAverage is clearly not working with this version.

I have kept all files including a personal log of what I did and when I did it. If anyone wants these files for reference please let me know.

2007-09-29, 11:17	#14
Brian-E "Brian" Jul 2007 The Netherlands CC5₁₆ Posts	Update Tested v25.5 for a couple of days running for 8 hours 30 minutes per day while CPUHours was set at 6. The file local.txt (as it is now called instead of local.ini) was rewritten several times but RollingAverage remained unchanged at 1000. This is certainly different from the behaviour of v24.14 which would have nudged RollingAverage up at an early stage. It's probably reasonable as well for the new version not to have changed RollingAverage yet because the program has not been running long enough to make a good estimate. Now I have decided to change behaviour and run v25.5 for only a short time each day or sometimes not at all to see if RollingAverage eventually gets adjusted below 1000. Will report results. I have one question/worry. Naturally it was important to run v25.5 in the same way as the way I was running v24.14 so that differences in behaviour could be confidently attributed to the different version of the software. However there is one fundamental difference: I am testing "dummy" exponents put into a file worktodo.txt without any communication with the server. I deliberately decided to do that because I didn't want to disrupt any genuine allocations of exponents when I'm only testing the new alpha software. And I thought that the calculation of RollingAverage should not be affected by this difference. But is that correct? Could it be that RollingAverage has remained at 1000 because the software is set up not to communicate with primenet? Does anyone know?

2007-10-09, 17:39	#19
Brian-E "Brian" Jul 2007 The Netherlands 7·467 Posts	update after 1 week Either the RollingAverage is not working in v25.5 or it is waiting until more data than the one week of testing that has so far been gathered before adjusting itself. During the last 7 days I have run LL testing for about 23 hours in total with CPUHours set at 6. RollingAverage should be about 500-600 on that basis but it has remained at 1000. The server was contacted several times with estimated completion dates, all apparently based on 6 hours per day running time. I will continue testing for a while with CPUHours=6 and real running time much less than that, and later I plan to change the test by setting CPUHours=1 and testing for significantly more than 1 hour per day. If anyone has their own testing suggestions please let me know. Brian.

2007-10-13, 10:31	#20
Brian-E "Brian" Jul 2007 The Netherlands 7×467 Posts	Testing stopped Tested a few more days with CPUHours set at 6 but running less than 1 hour per day. RollingAverage remained at 1000 and the server was often contacted with ETA dates still based on 6 hours per day. Today I decided to change over to running longer than the given CPUHours instead of shorter times but encountered the strange problem that when I tried to unreserve the exponent from the menu (I wanted to start completely as new with a new exponent so that the previous testing would hopefully not affect the new testing), after apparently succeeding the v25.5 then immediately requested new work from the server and was given... the same exponent as it had just unreserved! Further testing then continued on this exponent from where it left off! At this point I decided there isn't much point in continuing the test anyway because RollingAverage is clearly not working with this version. I have kept all files including a personal log of what I did and when I did it. If anyone wants these files for reference please let me know.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
How do I "override" the P4 effective equivalent or CPU rolling average	petrw1	PrimeNet	2	2013-06-17 05:17
Average Joe vs. Olympic Pros	MooooMoo	Lounge	28	2010-03-27 21:29

2007-09-25, 16:02	#12
garo Aug 2002 Termonfeckin, IE ACC₁₆ Posts	It sounds like RollingAverage is not being recalculated when the program runs for less hours than specified. That would explain why it is never revised downward except when the CPU is being shared by other processes. v25.5 is currently being alpha tested. Can you check if this behaviour also occurs with v25.5? The save files are not compatible so make sure you install v25.5 in another directory.

2007-09-30, 23:17	#17
garo Aug 2002 Termonfeckin, IE 5314₈ Posts	No don't ask for the secret line as that would get you TFs which again might not be so good for calculating the RollingAverage. Instead, communicate normally with primenet using v25.5. Then observe what happens over a week.