October 20, 2020, 08:24:29 am

Weird "time left"

Started by Ruskie, November 28, 2013, 06:18:22 am

Previous topic - Next topic

Ruskie

Hello everybody,

I have recently started with SETI@home, and I installed (yesterday) the optimized version for my i7 processor and GTX660 Nvidia card.
I used the installer so everything should have worked fine, however something strange is happening with the estimation of time left.
They are wildly higher than before!
For cuda50 tasks:
stock app processed them in about 15 minutes, and the estimation was 15 minutes right from the start
optimized version estimates 1h and something!
While processing estimated time decreases really fast, and in fact real computation either lasts around 6 minutes or 11 more or less (strangely enough, history shows this two duration alternates): so to give you a picture  it initally estimates let's say 1:10:00, then after 2 seconds 1:07:48, then after 2 other seconds 1:05:02 etc. etc. until the computation completes while estimated time still shows some minutes.

For MB tasks:
stock app estimated 5 hours or so.
when I restarted with the optimized app, the very same tasks started showing an estimates of 17 hours!
Here this estimates does not seem to drop dramatically as time goes by: it behaves as the estimation is good, decreasing most of the times for the same amount of elapsed seconds. In this case I still have to complete some WU 'cause BOINC switched to other projects, but judging by the already completed percentage I guess WU will still be faster than those computed with the stock app.

However, why the weird predicted times??

Thanks

Claggy

Found your host:

http://setiathome.berkeley.edu/show_host_detail.php?hostid=7151547

At present it doesn't have any completed tasks showing for any of the Anonymous platform planclasses:

http://setiathome.berkeley.edu/host_app_versions.php?hostid=7151547

Once each Anonymous platform planclass has completed '10 Number of tasks completed', then the Estimates will be close to right,
in the meantime they are either using the Stock app values, or totally unscaled values if the app version hasn't completed 10 completed tasks.

Claggy

Ruskie

Yep, this is the one where I installed the optimized app. Out of curiosity, how did you find it?
Anyway it's true as far as MB WU are concerned: as I said when they were halway through processing rosetta@home kicked in. But it's not true for cuda50, I have processed at least 30 or so...

By the way, why it has to be 10 workunits, how does the BOINC manager estimate time left?

Thanks!
Ruskie

Jason G

Quote from: Ruskie on November 28, 2013, 08:22:02 am
... how does the BOINC manager estimate time left?...


Very badly  :)
It is not the strongest of the species that survives, nor the most intelligent that survives. It is the one that is the most adaptable to change.
Charles Darwin
---
Chaos: When the present determines the future, but the approximate present does not approximately determine the future.
Edward Lorenz

Claggy

Quote from: Ruskie on November 28, 2013, 08:22:02 am
Yep, this is the one where I installed the optimized app. Out of curiosity, how did you find it?
Anyway it's true as far as MB WU are concerned: as I said when they were halway through processing rosetta@home kicked in. But it's not true for cuda50, I have processed at least 30 or so...

By the way, why it has to be 10 workunits, how does the BOINC manager estimate time left?

Thanks!
Ruskie
I used the user search and searched for 'Ruskie', saw one that had recently joined, and that he had the correct combination of CPU & GPU.

The server will scale the flops value for each app version once it's is on 11 completed tasks (So it's got an average for 10 completed 'full' tasks, ie it ignores any that exited early.)

You don't have any completed tasks for the 'SETI@home v7 (anonymous platform, CPU)' or the 'SETI@home v7 (anonymous platform, NVIDIA GPU)' app version yet,
the tasks you're completed would have been sent as Stock app versions, But then resent as Anonymous platform, ask for work again, and the  Anonymous platform app versions will show up:

Application details for host 7151547

Then complete and validate 10 full tasks.

Claggy

Ruskie

Quote from: Claggy on November 28, 2013, 08:44:27 am
You don't have any completed tasks for the 'SETI@home v7 (anonymous platform, CPU)' or the 'SETI@home v7 (anonymous platform, NVIDIA GPU)' app version yet,
the tasks you're completed would have been sent as Stock app versions, But then resent as Anonymous platform, ask for work again, and the  Anonymous platform app versions will show up:

Application details for host 7151547

Then complete and validate 10 full tasks.

Claggy


Thanks Claggy!
What's strange is that if I look for valid tasks I do see several for NVIDIA GPU: http://setiathome.berkeley.edu/results.php?userid=9948829&offset=0&show_names=0&state=4&appid= instead!
Anyway no problem, from what you say I understand that's a very well known "feature", nothing to worry about!

William

Quote from: Jason G on November 28, 2013, 08:33:30 am
Quote from: Ruskie on November 28, 2013, 08:22:02 am
... how does the BOINC manager estimate time left?...


Very badly  :)


fraction_done * fraction_done * frac_estimated + ( 1 - fraction_done * fraction_done) * wu_est

In numbers, when the task is at 50%

0.25 * estimate based on elapsed + 0.75 * initial estimate

The task needs to get to about 90% completion before the actual runtime governs the estimated time left.
MB tasks are not linear in their complettion - they speed up. So at 50% elapsed you already have more that 50% of total runtime.

The main thing that happens is that is screws up cache - I recommend to keep cache small and top up in little sips - something like 0.5-1.5 for the first (minium) value and 0.1 for the second ('additional') value.  That way you will get adjusted tasks as soon as the 11th validation has happened.
Under anonymous platform only new tasks are corrected for actual speed. So if you have a large bunch of overestimated ones it takes longer until you reach good values.
There are ways around that, but you need to fiddle with BOINC. Far easier to have a small cache in the first place - you can always make it larger after the estimates have come good.


Ouch. App details isn't updating. There are lots of validated tasks. Is that something that can be kicked, or is it better to get a clean slate and force a new host ID?
Remember it's turkey weekend in the US. We won't get hold of either Eric to use a crowbar on the database or David to debug - well to report, thugh I think we've already done that a few times...

Under the circumstances it might be best to force a new host ID now and deploy the rolling pin later?

Ruskie, care to perform surgery on an open BOINC? I doubt it will sort itself out on it's own.
A person who won't read has no advantage over one who can't read. (Mark Twain)

Claggy

Quote from: William on November 28, 2013, 07:50:18 pm
Ouch. App details isn't updating. There are lots of validated tasks. Is that something that can be kicked, or is it better to get a clean slate and force a new host ID?
Remember it's turkey weekend in the US. We won't get hold of either Eric to use a crowbar on the database or David to debug - well to report, thugh I think we've already done that a few times...

He's now got the anonymous platform app version entries, 2 completed for v7 CPU, and 3 completed for v7 Nvidia GPU.

Claggy

Ruskie

I wouldn't mind if you tell me what to do; I can do with computers but are a total newbie as far as S@H hacking is concerned.
However from last Claggy post I'd still say it really is still all right, is it?

William

Quote from: Ruskie on November 29, 2013, 01:01:43 am
I wouldn't mind if you tell me what to do; I can do with computers but are a total newbie as far as S@H hacking is concerned.
However from last Claggy post I'd still say it really is still all right, is it?


If it is updating now, it should be ok - you'll have to wait for those 11 validations before you have good estimates, but now that is just a matter of patience.

I didn't expect it to get good without intervention ::) - sorry.

Have fun crunching and welcome to Seti and CA!
A person who won't read has no advantage over one who can't read. (Mark Twain)

Ruskie

Quote from: William on November 29, 2013, 01:11:39 am
Have fun crunching and welcome to Seti and CA!


Thank you! It seems to me that nvidia tasks are starting to get it right now. CPU tasks are still way off, but they're not running lately 'cause rosetta has taken slots over. We'll see!

Powered by EzPortal