Panic Mode On (111) Server Problems?

Message boards : Number crunching : Panic Mode On (111) Server Problems?
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 27 · 28 · 29 · 30 · 31 · Next

AuthorMessage
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1930734 - Posted: 18 Apr 2018, 20:13:04 UTC - in response to Message 1930718.  
Last modified: 18 Apr 2018, 20:40:25 UTC

Yes, everything is running smoothly now.
My only gripe atm is this :

Run time CPU time Credit Application
913.50 902.64 447.46 AstroPulse v7 v7.10 (opencl_nvidia_100) windows_intelx86
18,403.52 18,136.26 395.58 AstroPulse v7 v7.03 (sse2) windows_x86_64

Hardly fair.

Without a link to the tasks cited, it is impossible to say why the difference. Most likely the amount of blanking was different between the tasks. The higher the blanking, the longer to compute and the higher the credit awarded.
I'm seeing results that seem to imply AP tasks run against CPUs award much lower credit than those run by Two GPUs. There is a bit of blanking, but at first look it seems some of the longer CPU tasks also pay less credit. I brought my ATI cards out of retirement just to run some APs, but it seems they are getting robbed. The blanked tasks take longer on the ATIs, but if the CPU takes a short time then the ATIs get much less credit for a longer task. Take a look at some of the run-times verses credit, https://setiathome.berkeley.edu/results.php?hostid=6796475&offset=60&appid=20
This one is more depressing than most, it's a longer than normal time, but scores a faction of normal;
  Task    Computer	        Sent	             Time reported                 Status                Run time   CPU time    Credit	                 Application
6573740896 8260247   18 Apr 2018, 1:17:28 UTC  18 Apr 2018, 7:47:00 UTC   Completed and validated 	14,620.79   14,551.17   163.35    AstroPulse v7 v7.03 (sse2) windows_x86_64
6573740897 6796475   18 Apr 2018, 1:17:28 UTC  18 Apr 2018, 6:51:54 UTC   Completed and validated        2,132.81      441.49   163.35    AstroPulse v7 Anonymous platform (ATI GPU)

Blah! 131.04 pts for a normal AP, https://setiathome.berkeley.edu/workunit.php?wuid=2941091313
ID: 1930734 · 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: 22532
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1930735 - Posted: 18 Apr 2018, 20:27:19 UTC

One of the (major) deficiencies of CreditScrew is that is will award to lower calculated score to both parties. And it doesn't play well with GPUs in general, or with GPUs doing multiple tasks in particular.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1930735 · Report as offensive
juan BFP Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 16 Mar 07
Posts: 9786
Credit: 572,710,851
RAC: 3,799
Panama
Message 1930739 - Posted: 18 Apr 2018, 20:38:39 UTC - in response to Message 1930730.  

In your, mine and allmost all opinion but not in the opinion of CreditScrew.

Oh no, he mentioned the unmentionable. Now I need to take a bath in holy water :-)


ID: 1930739 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1930741 - Posted: 18 Apr 2018, 20:41:15 UTC

I'm sure there is a good reason why one was assigned to the CPU (SSE2) while the other was assigned to the GPU (opencl), but still.. !

I can't seem to grab those tasks to get a look a their estimated flops. Would be one way to point at what credit should be awarded.

Credit is handled by the CreditNew award mechanism which is admittedly flawed as most will attest. It was based originally on the number of Cobblestones used to compute a task. The same amount of cobblestones would be used to compute the same task whether it was computed on a gpu or cpu. The gpu is the more powerful computer and so would run the task in a shorter wall clock time compared to the slower cpu which would need a much longer amount of wall clock time to accumulate the same number of Cobblestones.

That base mechanism was the original start of CreditNew. But then CreditNew got corrupted about 8 years ago by the modification of the award mechanism based on theoretical AVX compute profiling.

And it has been debased to CreditScrew ever since. Can't do anything about it so just accept it.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1930741 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1930742 - Posted: 18 Apr 2018, 21:01:36 UTC

Seems there is also a hit to the MB Arecibo tasks Credit. My one Mac has been running mostly Arecibo Non-APs tasks for the past few days. Usually when that happens it will be up in the Mid-90k range, However, it's currently in the Lower 80k range. The only reason that could be, would be the Tons of VLARs it's been running. So as usual, any change to Credit awarded is Negative. It's ALWAYS Negative. My Theory is CreditFew is actually designed to consistently award Fewer and Fewer awards, it certainly preforms that way anyway.
ID: 1930742 · Report as offensive
Profile Bill G Special Project $75 donor
Avatar

Send message
Joined: 1 Jun 01
Posts: 1282
Credit: 187,688,550
RAC: 182
United States
Message 1930744 - Posted: 18 Apr 2018, 21:13:27 UTC - in response to Message 1930742.  
Last modified: 18 Apr 2018, 21:22:35 UTC

With regard to the length of time it takes to do an AP:
As a general rule when run on the CPU instead of the GPU it takes about 7 times longer. They both will give the same amount of credit it seems.

The only way that I know of around that is to not run APs on the CPU.

(I think the the CPU credit is just about the same as non-ap work despite the time difference of AP to non-AP work. Whithout doing massive amounts of data averaging that is my take on it.)

SETI@home classic workunits 4,019
SETI@home classic CPU time 34,348 hours
ID: 1930744 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1930745 - Posted: 18 Apr 2018, 21:14:22 UTC - in response to Message 1930742.  

Seems there is also a hit to the MB Arecibo tasks Credit. My one Mac has been running mostly Arecibo Non-APs tasks for the past few days. Usually when that happens it will be up in the Mid-90k range, However, it's currently in the Lower 80k range. The only reason that could be, would be the Tons of VLARs it's been running. So as usual, any change to Credit awarded is Negative. It's ALWAYS Negative. My Theory is CreditFew is actually designed to consistently award Fewer and Fewer awards, it certainly preforms that way anyway.


. . Exactly, any change in work type and the credit awarded across the board will decrease ....

. . I will say however that the decline with the Arecibo VLARs on nvidia GPUs is slower than I had expected. Slow but inexorable.

Stephen

:(
ID: 1930745 · Report as offensive
juan BFP Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 16 Mar 07
Posts: 9786
Credit: 572,710,851
RAC: 3,799
Panama
Message 1930746 - Posted: 18 Apr 2018, 21:21:05 UTC - in response to Message 1930742.  

So as usual, any change to Credit awarded is Negative.

To try to mantain our RAC we need to put to work more and more powerfull hosts.
Maybe that's the hidden idea behind CreditScrew (or CreditFew)
My +/- 190K/host is now producing <160K/day.
ID: 1930746 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14679
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1930748 - Posted: 18 Apr 2018, 21:29:08 UTC - in response to Message 1930723.  

No matter how the tasks are built, one that runs for over 5hrs should pay more than one that takes 15 minutes, in my opinion.
A workunit is a fixed amount of work. Faster hardware (GPU) will complete it more quickly than slower hardware (CPU). I don't see anything wrong with that - the credit is awarded for the work (see definition of cobblestone).
ID: 1930748 · Report as offensive
Profile Jord
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 15184
Credit: 4,362,181
RAC: 3
Netherlands
Message 1930750 - Posted: 18 Apr 2018, 21:35:47 UTC

So I guess this isn't good for getting data from Arecibo either: Puerto Rico is Once Again Hit by an Islandwide Blackout

An electrical contractor working to restore power in Puerto Rico accidentally knocked out a major transmission line on Wednesday, leaving the entire island without power nearly seven months after Hurricane Maria destroyed the electrical grid.

It could take up to 36 hours to restore electricity to nearly 1.5 million affected customers, the Puerto Rico Electric Power Authority said.

ID: 1930750 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14679
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1930755 - Posted: 18 Apr 2018, 22:07:06 UTC - in response to Message 1930750.  

I thought the last we heard, the Arecibo radio telescope was limping along on fossil (diesel) generators, independently of the island's civilian electricity network? I'll leave you to think about whether scientists should isolate themselves like that, but it happens.
ID: 1930755 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1930757 - Posted: 18 Apr 2018, 22:29:25 UTC - in response to Message 1930744.  

With regard to the length of time it takes to do an AP:
As a general rule when run on the CPU instead of the GPU it takes about 7 times longer. They both will give the same amount of credit it seems.

The only way that I know of around that is to not run APs on the CPU.

(I think the the CPU credit is just about the same as non-ap work despite the time difference of AP to non-AP work. Whithout doing massive amounts of data averaging that is my take on it.)

I've been rescheduling any AP CPU tasks to GPU simply because of the efficiency gain. The side benefit it seems is that the scheduler now thinks the CPU is the most efficient compute device and it seems it has been loading me up with tons of AP CPU tasks. Which only means more AP tasks for the gpus and more credit. I have had upwards of 30+ AP tasks on all hosts now for a couple of weeks at all times.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1930757 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1930762 - Posted: 18 Apr 2018, 22:53:49 UTC - in response to Message 1930757.  

Have you run the math on the machines running the Special App? I think you will find the Faster GPUs actually Lose Credit when running APs. They produce More Credit by running normal Arecibo MBs than APs. The VLARs are about even. So, you are going thru a lot of trouble, and destroying your APR, for nothing. I've long recommended the Faster GPUs running the Special App Avoid APs, and leave them for the slower machines. You'll notice my fastest machine doesn't run APs.
ID: 1930762 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1930764 - Posted: 18 Apr 2018, 23:05:16 UTC - in response to Message 1930762.  

Have you run the math on the machines running the Special App? I think you will find the Faster GPUs actually Lose Credit when running APs. They produce More Credit by running normal Arecibo MBs than APs. The VLARs are about even. So, you are going thru a lot of trouble, and destroying your APR, for nothing. I've long recommended the Faster GPUs running the Special App Avoid APs, and leave them for the slower machines. You'll notice my fastest machine doesn't run APs.

All this may be true. But I just have a big bother with inefficiency. It just bugs me to see a cpu core tied up for 4 hours with a AP task when it could be done in 12 minutes on the gpu. As you say, it likely is hurting my RAC. And it definitely is messing with the cpu APR.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1930764 · Report as offensive
Profile betreger Project Donor
Avatar

Send message
Joined: 29 Jun 99
Posts: 11416
Credit: 29,581,041
RAC: 66
United States
Message 1930767 - Posted: 18 Apr 2018, 23:55:17 UTC

IMO if I cared about credit I wouldn't run Seti at all. If you want credit there are projects such as Collatz that award insane credit. That is aside of my opinion that trying to disprove what seems to be a logical regression by brute force is also a waste of electricity.
ID: 1930767 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1930771 - Posted: 19 Apr 2018, 0:05:25 UTC - in response to Message 1930748.  

No matter how the tasks are built, one that runs for over 5hrs should pay more than one that takes 15 minutes, in my opinion.
A workunit is a fixed amount of work. Faster hardware (GPU) will complete it more quickly than slower hardware (CPU). I don't see anything wrong with that - the credit is awarded for the work (see definition of cobblestone).


. . Sadly Richard, while what you say is true that credit is supposed to represent the actual work done, in reality there is little to no consistency and certainly not any connection to the defined cobblestone ...

Stephen

:(
ID: 1930771 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1930772 - Posted: 19 Apr 2018, 0:09:46 UTC - in response to Message 1930762.  
Last modified: 19 Apr 2018, 0:13:02 UTC

Have you run the math on the machines running the Special App? I think you will find the Faster GPUs actually Lose Credit when running APs. They produce More Credit by running normal Arecibo MBs than APs. The VLARs are about even. So, you are going thru a lot of trouble, and destroying your APR, for nothing. I've long recommended the Faster GPUs running the Special App Avoid APs, and leave them for the slower machines. You'll notice my fastest machine doesn't run APs.


. . Yep, exactly what I have found, on good GPUs running a special sauce form of CUDA you do better with tasks other than AP. But Keith is also running Windows boxes and I think that is where he is benefitting.

Stephen

. .
ID: 1930772 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1930773 - Posted: 19 Apr 2018, 0:11:49 UTC - in response to Message 1930767.  

IMO if I cared about credit I wouldn't run Seti at all. If you want credit there are projects such as Collatz that award insane credit. That is aside of my opinion that trying to disprove what seems to be a logical regression by brute force is also a waste of electricity.


. . Well there's Einstein@home, they give good credit and are doing useful work.

Stephen

.
ID: 1930773 · Report as offensive
Sirius B Project Donor
Volunteer tester
Avatar

Send message
Joined: 26 Dec 00
Posts: 24912
Credit: 3,081,182
RAC: 7
Ireland
Message 1930776 - Posted: 19 Apr 2018, 0:27:55 UTC

Ah but everyone wants to be the 1st in finding E.T. so they can tell him to phone home :-)
ID: 1930776 · Report as offensive
Profile betreger Project Donor
Avatar

Send message
Joined: 29 Jun 99
Posts: 11416
Credit: 29,581,041
RAC: 66
United States
Message 1930777 - Posted: 19 Apr 2018, 0:30:11 UTC - in response to Message 1930773.  

. . Well there's Einstein@home, they give good credit and are doing useful work.

Yep I'm there and maintain a RAC ~500K, as an aside a couple of days ago Richard posted that he thought E@H probably was truest to the cobblestone credit issue but I don't crunch there for the credit.
ID: 1930777 · Report as offensive
Previous · 1 . . . 27 · 28 · 29 · 30 · 31 · Next

Message boards : Number crunching : Panic Mode On (111) Server Problems?


 
©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.