That wierd

Message boards : Number crunching : That wierd
Message board moderation

To post messages, you must log in.

AuthorMessage
W-K 666 Project Donor
Volunteer tester

Send message
Joined: 18 May 99
Posts: 19012
Credit: 40,757,560
RAC: 67
United Kingdom
Message 1154489 - Posted: 21 Sep 2011, 14:17:03 UTC
Last modified: 21 Sep 2011, 14:18:33 UTC

All tasks received with the last three hours, have the same estimated completion time, whether they are for the cpu or the gpu.

They are all MB's and with DCF at 0.93 they are reading 02:33:00 +/- 3 mins
ID: 1154489 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22160
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1154494 - Posted: 21 Sep 2011, 14:22:48 UTC

Hmm - I've just looked at my load, that have come in since this morning (UK).
All have the same time (CPU & GPU mix), but a different one to you - 2hrs16min....
Obviously using dice to determine the run time, which might be better if they could get the correct sided dice...
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1154494 · Report as offensive
W-K 666 Project Donor
Volunteer tester

Send message
Joined: 18 May 99
Posts: 19012
Credit: 40,757,560
RAC: 67
United Kingdom
Message 1154499 - Posted: 21 Sep 2011, 14:36:13 UTC

Now that computer has started crunching theese tasks d/loaded recently I can say they are all VHAR's the "tapes" are all dated 02, 21, 24jl11
ID: 1154499 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22160
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1154502 - Posted: 21 Sep 2011, 14:41:40 UTC

That may be the case, but it doesn't explain why the "guess time" for both CPU & GPU tasks are the same.
Also the CPU guess time is more or less correct, but the GPU guess overestimates by a factor of about 40.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1154502 · Report as offensive
W-K 666 Project Donor
Volunteer tester

Send message
Joined: 18 May 99
Posts: 19012
Credit: 40,757,560
RAC: 67
United Kingdom
Message 1154503 - Posted: 21 Sep 2011, 14:48:30 UTC - in response to Message 1154502.  

That may be the case, but it doesn't explain why the "guess time" for both CPU & GPU tasks are the same.
Also the CPU guess time is more or less correct, but the GPU guess overestimates by a factor of about 40.

The times on my cpu are ~40mins, and on the GPU ~8mins. Which is normal time for VHAR tasks.
So if my DCF was 1.00000 then the estimated times are about 4 * for the CPU, and 20 * for the GPU.
ID: 1154503 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1154506 - Posted: 21 Sep 2011, 15:01:26 UTC - in response to Message 1154503.  

That may be the case, but it doesn't explain why the "guess time" for both CPU & GPU tasks are the same.
Also the CPU guess time is more or less correct, but the GPU guess overestimates by a factor of about 40.

The times on my cpu are ~40mins, and on the GPU ~8mins. Which is normal time for VHAR tasks.
So if my DCF was 1.00000 then the estimated times are about 4 * for the CPU, and 20 * for the GPU.

You can always get a quick 'rule of thumb' on task duration by looking at the deadline column - shorties are currently arriving with a deadline of somewhat after 09:00 UTC 05 Oct 2011 - anything in late October or early November is a mid- or low- AR task.

You will get identical runtime estimates for CPU and GPU tasks if: you are running anonymous platform, and no APR correction is currently being applied to the GPU tasks on that host.

The could mean that your individual computer has picked up a new HostID along the way, and started a fresh set of application details: or it could mean that APR has been disabled for the whole project, for the time being at least.

Could you have a delve into your host records, please, and help us work out which it is?
ID: 1154506 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22160
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1154508 - Posted: 21 Sep 2011, 15:09:27 UTC - in response to Message 1154506.  

Looking at the computer ID suggests that it hasn't changed since 2009, so it looks as if APR is not functioning.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1154508 · Report as offensive
W-K 666 Project Donor
Volunteer tester

Send message
Joined: 18 May 99
Posts: 19012
Credit: 40,757,560
RAC: 67
United Kingdom
Message 1154511 - Posted: 21 Sep 2011, 15:23:41 UTC - in response to Message 1154506.  
Last modified: 21 Sep 2011, 15:28:45 UTC

That may be the case, but it doesn't explain why the "guess time" for both CPU & GPU tasks are the same.
Also the CPU guess time is more or less correct, but the GPU guess overestimates by a factor of about 40.

The times on my cpu are ~40mins, and on the GPU ~8mins. Which is normal time for VHAR tasks.
So if my DCF was 1.00000 then the estimated times are about 4 * for the CPU, and 20 * for the GPU.

You can always get a quick 'rule of thumb' on task duration by looking at the deadline column - shorties are currently arriving with a deadline of somewhat after 09:00 UTC 05 Oct 2011 - anything in late October or early November is a mid- or low- AR task.

You will get identical runtime estimates for CPU and GPU tasks if: you are running anonymous platform, and no APR correction is currently being applied to the GPU tasks on that host.

The could mean that your individual computer has picked up a new HostID along the way, and started a fresh set of application details: or it could mean that APR has been disabled for the whole project, for the time being at least.

Could you have a delve into your host records, please, and help us work out which it is?

HostID hasn't changed and APR correction
SETI@home Enhanced (anonymous platform, CPU)
Number of tasks completed	1624
Max tasks per day	214
Number of tasks today	21
Consecutive valid tasks	114
Average processing rate	21.008039955223
Average turnaround time	4.26 days

SETI@home Enhanced (anonymous platform, nvidia GPU)
Number of tasks completed	2716
Max tasks per day	454
Number of tasks today	52
Consecutive valid tasks	355
Average processing rate	104.62213864253
Average turnaround time	2.00 days

Astropulse v505 (anonymous platform, CPU)
Number of tasks completed	64
Max tasks per day	126
Number of tasks today	0
Consecutive valid tasks	27
Average processing rate	55.896231237772
Average turnaround time	2.22 days
ID: 1154511 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1154518 - Posted: 21 Sep 2011, 15:36:13 UTC - in response to Message 1154511.  

OK, I've got a host I'm monitoring very closely for straws in the wind. It runs VHARs in 4 minutes 15 seconds, give or take, and runs with a DCF of about 0.025 when showing accurate estimates for continuous shorties - just high enough to allow continuous work fetch.

That's with the doubled APR cap introduced last week. Before that was released, DCF on continual VHARs was below the magic 0.02, so I had to resort to manual intervention to get work.

I've just had a mid-AR resend go through, so DCF has been kicked higher - 0.0343 - and VHAR estimates are now 05:46. That sounds proportionate, suggesting the APR treatment is unchanged since last week. Anyone else got any observations to chip in?
ID: 1154518 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22160
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1154525 - Posted: 21 Sep 2011, 15:50:36 UTC

Richard - details for your consideration
(this host takes about 4mins to complete a "shorty" on the GPU - a GTS250)

SETI@home Enhanced 6.03 windows_intelx86
Number of tasks completed 345
Max tasks per day 424
Number of tasks today 0
Consecutive valid tasks 201
Average processing rate 4.2641769609151
Average turnaround time 7.66 days

SETI@home Enhanced 6.09 windows_intelx86 (cuda23)
Number of tasks completed 1371
Max tasks per day 301
Number of tasks today 0
Consecutive valid tasks 6
Average processing rate 25.48554602749
Average turnaround time 10.43 days

SETI@home Enhanced (anonymous platform, CPU)
Number of tasks completed 830
Max tasks per day 677
Number of tasks today 0
Consecutive valid tasks 578
Average processing rate 4.67252804848
Average turnaround time 6.44 days

SETI@home Enhanced (anonymous platform, nvidia GPU)
Number of tasks completed 13063
Max tasks per day 1681
Number of tasks today 181
Consecutive valid tasks 1581
Average processing rate 193.10907358688
Average turnaround time 1.27 days


Astropulse v505 (anonymous platform, CPU)
Number of tasks completed 54
Max tasks per day 127
Number of tasks today 0
Consecutive valid tasks 28
Average processing rate 7.0024699071515
Average turnaround time 7.92 days
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1154525 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1154535 - Posted: 21 Sep 2011, 15:57:53 UTC - in response to Message 1154525.  

And what's your DCF, when estimates are matching reality? (after a long continuous run of VHAR 'shorties' on GPU is the best time to check).

I'm watching a week-old 9800GT account, with

SETI@home Enhanced (anonymous platform, nvidia GPU)
Number of tasks completed 582
Max tasks per day 338
Number of tasks today 60
Consecutive valid tasks 238
Average processing rate 191.29744959377
Average turnaround time 0.58 days

- so pretty well matched to your GTS250
ID: 1154535 · Report as offensive
Iona
Avatar

Send message
Joined: 12 Jul 07
Posts: 790
Credit: 22,438,118
RAC: 0
United Kingdom
Message 1154572 - Posted: 21 Sep 2011, 17:03:07 UTC

I have also been getting GPU tasks sent which have estimated times very similar to CPU tasks. I strongly suspect you are correct, Richard, when you opine that APR has been disabled.....my ludicrously high ATI GPU MB APR has not changed a bit. That being so, I am relieved that I correctly saw the chance to process GPU MB, with last weeks very high time estimates - it did me a favour. Usually, a 'shortie' takes around 24 mins to the GPUs 12 mins, but with AP, things are a bit different.....roughly 10 hrs CPU to the GPUs 3 hrs. Application details are as below.

SETI@home Enhanced (anonymous platform, CPU)
Number of tasks completed 715
Max tasks per day 2984
Number of tasks today 0
Consecutive valid tasks 2364
Average processing rate 28.023010646597
Average turnaround time 0.94 days

SETI@home Enhanced (anonymous platform, ATI GPU)
Number of tasks completed 15
Max tasks per day 139
Number of tasks today 0
Consecutive valid tasks 40
Average processing rate 1390.280653876
Average turnaround time 0.10 days

Astropulse v505 (anonymous platform, CPU)
Number of tasks completed 24
Max tasks per day 164
Number of tasks today 0
Consecutive valid tasks 64
Average processing rate 44.213757313418
Average turnaround time 2.72 days

Astropulse v505 (anonymous platform, ATI GPU)
Number of tasks completed 22
Max tasks per day 126
Number of tasks today 0
Consecutive valid tasks 26
Average processing rate 193.72450987751
Average turnaround time 1.38 days

Task duration correction factor 0.207809




Don't take life too seriously, as you'll never come out of it alive!
ID: 1154572 · Report as offensive
Josef W. Segur
Volunteer developer
Volunteer tester

Send message
Joined: 30 Oct 99
Posts: 4504
Credit: 1,414,761
RAC: 0
United States
Message 1154673 - Posted: 21 Sep 2011, 21:54:38 UTC

As it happens, with <avg_cpus>0.04</avg_cpus> and <count>1</count> in the app_version section of an app_info.xml file without <flops> the BOINC core client will be sending a <flops> value for the GPU which is just 4% higher than the <flops> it sends for CPU. The server scaling is based on twice those values if the APRs are above that. The core client's DCF affects each project task equally, so it is not surprising if CPU and GPU runtime estimates are nearly equal in that case.

A host with 1.92589e09 Whetstones and MB APRs for anonymous platform of 4.67252804848 CPU and 193.10907358688 GPU is in that condition.

So is a host with 3.07063e09 Whetstones, MB APRs of 28.023010646597 and 1390.280653876, and AP APRs of 44.213757313418 and 193.72450987751.
                                                                   Joe
ID: 1154673 · Report as offensive
Profile Fred J. Verster
Volunteer tester
Avatar

Send message
Joined: 21 Apr 04
Posts: 3252
Credit: 31,903,643
RAC: 0
Netherlands
Message 1154691 - Posted: 21 Sep 2011, 22:18:02 UTC - in response to Message 1154673.  
Last modified: 21 Sep 2011, 22:27:39 UTC

Another host, with a GTS250 and a Q6600, stock, WIN32XP,
this
host.


DETAILS
SETI@home Enhanced (anonymous platform, Unknone)
Number of tasks completed 60
Max tasks per day 60
Number of tasks today 0
Consecutive valid tasks 0
Average processing rate 19.287435425713
Average Turnaround Time 3.49 days
SETI@home Enhanced (anonymous platform, CPU)
Number of tasks completed 2003
Max tasks per day 187
Number of tasks today 0
Consecutive valid tasks 98
Average processing rate 18.342186573779
Average Turnaround Time 2.65 days
SETI@home Enhanced (anonymous platform, nvidia GPU)
Number of tasks completed 28189
Max tasks per day 1643
Number of tasks today 309
Consecutive valid tasks 1544
Average processing rate 177.2357327562
Average Turnaround Time 1.66 days
Astropulse v505 (anonymous platform, CPU)
Number of tasks completed 54
Max tasks per day 104
Number of tasks today 0
Consecutive valid tasks 4
Average processing rate 38.590850210474
Average Turaround Time 4.58 days


But only this host got many shorties, all other MBs are being crunched by
(SSSE3;optimized) CPUs, MW is blowing my 5870 GPUs to it's limit, but the new
(Bêta)Rev.365 works very good and has a far better GPU load, first GPU=44% avg., second 36% avg.
(Highest was 57% on 1st GPU and 51% for the 2nd GPU,
both average over a ~1000 second MB WU).
But that's
this host
and drifting away.......................
ID: 1154691 · Report as offensive
Cosmic_Ocean
Avatar

Send message
Joined: 23 Dec 00
Posts: 3027
Credit: 13,516,867
RAC: 13
United States
Message 1154694 - Posted: 21 Sep 2011, 22:23:09 UTC

What I have observed is that on separate work requests, as long as I have not completed and reported any tasks, I will get identical ETAs on all WUs that come in, until at least one gets reported. On my main cruncher, this can be upwards of a full 24 hours in which I can get new WUs with identical ETAs.
Linux laptop:
record uptime: 1511d 20h 19m (ended due to the power brick giving-up)
ID: 1154694 · Report as offensive
Profile Fred J. Verster
Volunteer tester
Avatar

Send message
Joined: 21 Apr 04
Posts: 3252
Credit: 31,903,643
RAC: 0
Netherlands
Message 1154721 - Posted: 21 Sep 2011, 23:05:51 UTC - in response to Message 1154694.  
Last modified: 21 Sep 2011, 23:09:28 UTC

Isn't that part of the way BOINC is designed to work, also after some time, and a
sufficient number of completed and validated work is uploaded and canonical result is reached, after awhile estimated time, should not be that far off.
That's ofcoarse pure theory cause if it's attached network hasn't sufficient bandwidth,
it's gettin further and further delays, up-- &down-loads get slowed or stuck!

(I do notice a speedup in downloads, though).
ID: 1154721 · Report as offensive

Message boards : Number crunching : That wierd


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.