Welcome to MilkyWay@home

Separation Application Shutting Down on Tuesday, Jun 20th

Message boards : News : Separation Application Shutting Down on Tuesday, Jun 20th
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 8 · 9 · 10 · 11 · 12 · 13 · 14 . . . 16 · Next

AuthorMessage
Link
Avatar

Send message
Joined: 19 Jul 10
Posts: 589
Credit: 18,930,552
RAC: 4,413
Message 75994 - Posted: 27 Jun 2023, 14:43:53 UTC - in response to Message 75988.  
Last modified: 27 Jun 2023, 14:46:36 UTC

I don't understand, how is it calculating your credits? If we are both given 1 identical task and run it on the same hardware, but you optimise your code, do we get the same credit per task because it's on the same card, or do we get different amounts because we took different lengths of time?
Yes, we get both same credit for that particular task, but we get less that we should for that amount of work, because CreditNew thinks the task was smaller and not that we used our hardware more efficiently.
ID: 75994 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Link
Avatar

Send message
Joined: 19 Jul 10
Posts: 589
Credit: 18,930,552
RAC: 4,413
Message 75995 - Posted: 27 Jun 2023, 14:52:37 UTC - in response to Message 75993.  

Thanks, everyone for the help. One of the 2 slow machines has a 64 bit operating system and the other one has a 32 bit operating system. From what I understand the 64 bit unit will run N Body (very slowly), but the 32 bit computer won’t? Even though I run a lot of machines, I don’t know a whole lot about them.
Yes, you need 64bit for Nbody. There are other astrophysics projects which work on 32bit, someone will add a list in a moment. I'm not sure how to tell.
Some projects were posted in this thread not long ago.
ID: 75995 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mr P Hucker
Avatar

Send message
Joined: 5 Jul 11
Posts: 990
Credit: 376,143,149
RAC: 5
Message 75996 - Posted: 27 Jun 2023, 15:02:36 UTC - in response to Message 75994.  

I don't understand, how is it calculating your credits? If we are both given 1 identical task and run it on the same hardware, but you optimise your code, do we get the same credit per task because it's on the same card, or do we get different amounts because we took different lengths of time?
Yes, we get both same credit for that particular task, but we get less that we should for that amount of work, because CreditNew thinks the task was smaller and not that we used our hardware more efficiently.
Doesn't credit new work it out on the basis of how long they take? Well if it's possible to do those tasks in less time buy running the hardware efficiently, and it's not possible to do so on another project, then you should get less credit on the first project, because you've spent less time on it.

Or, it could be based on pure flops, but that would mean you got less credit for something more difficult per flop, but more useful results.

Or, well there's billions of ways of assessing the points. You just can't compare apples and oranges. Am I better than you if I build a hospital and you build a university? I've cured people but you've invented things. Credits across projects means nothing.
The above was double spaced between sentences, I apologise for the forum software ruining my post.
ID: 75996 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Link
Avatar

Send message
Joined: 19 Jul 10
Posts: 589
Credit: 18,930,552
RAC: 4,413
Message 75999 - Posted: 27 Jun 2023, 17:29:58 UTC - in response to Message 75996.  

Doesn't credit new work it out on the basis of how long they take?
Yes, it does, and that's the issue as optimizations of the app or the hardware (which are not detected by the very basic BOINC benchmark) are not taken into account.


Well if it's possible to do those tasks in less time buy running the hardware efficiently, and it's not possible to do so on another project, then you should get less credit on the first project, because you've spent less time on it.
The credits do not show how much time someone has spent on a project, but how much computation that person has contributed to the project in "Cobblestones", which can be translated FLOPs or integer operations (whetstone/dhrystone). That's their definition, not time spent on an application.

Now if someone uses an optimized app, which needs for some computation just half the cycles an unoptimized app would need (for example), than CreditNew doesn't know at first wether the WU needed just half amount of computation or if that computer is faster than according to the benchmark values, but it will find it out by comparing with other computers. That works however only as long as only few use optimized apps and the stock apps are not optimized at all. When more optimized apps are used and even the stock apps get optimized as well, CreditNew does no longer assume, that the computers are faster than according to benchmark, but that the WUs contain less work. That's the reason, why over the time credit for a SETI WU was getting lower and lower.


Credits across projects means nothing.
But at least within one project and one app they should be comparable and not decrease for the same amount of work as applications and hardware get better. If that happens, there's clearly something wrong.
ID: 75999 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 8 May 09
Posts: 3319
Credit: 520,377,705
RAC: 23,403
Message 76006 - Posted: 27 Jun 2023, 23:17:51 UTC - in response to Message 75996.  

Mr P Hucker posted

Credits across projects means nothing.


But it's VERY hard to teach that to new people or to people who don't run alot of projects, they just don't understand why project A gives out X credits per task and project B gives out Z credits per task and why they can't be the same.
ID: 76006 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mr P Hucker
Avatar

Send message
Joined: 5 Jul 11
Posts: 990
Credit: 376,143,149
RAC: 5
Message 76012 - Posted: 28 Jun 2023, 8:56:02 UTC - in response to Message 75999.  

Now if someone uses an optimized app, which needs for some computation just half the cycles an unoptimized app would need (for example), than CreditNew doesn't know at first wether the WU needed just half amount of computation or if that computer is faster than according to the benchmark values, but it will find it out by comparing with other computers. That works however only as long as only few use optimized apps and the stock apps are not optimized at all. When more optimized apps are used and even the stock apps get optimized as well, CreditNew does no longer assume, that the computers are faster than according to benchmark, but that the WUs contain less work. That's the reason, why over the time credit for a SETI WU was getting lower and lower.

Credits across projects means nothing.
But at least within one project and one app they should be comparable and not decrease for the same amount of work as applications and hardware get better. If that happens, there's clearly something wrong.
Am I missing something? Why not just count the number of flops and call 1000 flops a credit?
The above was double spaced between sentences, I apologise for the forum software ruining my post.
ID: 76012 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mr P Hucker
Avatar

Send message
Joined: 5 Jul 11
Posts: 990
Credit: 376,143,149
RAC: 5
Message 76013 - Posted: 28 Jun 2023, 8:58:33 UTC - in response to Message 76006.  

But it's VERY hard to teach that to new people or to people who don't run alot of projects, they just don't understand why project A gives out X credits per task and project B gives out Z credits per task and why they can't be the same.
I don't see why anyone cares about cross-project credits. If I'm competing against you for example, I'll run the same project and try to catch up with you. Saying I've done 4 million credits on MW and you've only done 3 million on Einstein doesn't make sense. It would be like me saying I ran a certain distance and you cycled a certain distance. Different activities, not comparable.
The above was double spaced between sentences, I apologise for the forum software ruining my post.
ID: 76013 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 8 May 09
Posts: 3319
Credit: 520,377,705
RAC: 23,403
Message 76014 - Posted: 28 Jun 2023, 9:17:22 UTC - in response to Message 76013.  

But it's VERY hard to teach that to new people or to people who don't run alot of projects, they just don't understand why project A gives out X credits per task and project B gives out Z credits per task and why they can't be the same.


I don't see why anyone cares about cross-project credits. If I'm competing against you for example, I'll run the same project and try to catch up with you. Saying I've done 4 million credits on MW and you've only done 3 million on Einstein doesn't make sense. It would be like me saying I ran a certain distance and you cycled a certain distance. Different activities, not comparable.


But new and some old people don't see it that way, they see a credit as a credit as a credit. Kinda like electricity you have volts and watts and we in the US have volts and watts, you have mm's, cm's and meters while we have inches, feet and yards but they are comparable.
ID: 76014 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 8 May 09
Posts: 3319
Credit: 520,377,705
RAC: 23,403
Message 76015 - Posted: 28 Jun 2023, 9:23:23 UTC - in response to Message 76012.  

Credits across projects means nothing.
But at least within one project and one app they should be comparable and not decrease for the same amount of work as applications and hardware get better. If that happens, there's clearly something wrong.[/quote]

Am I missing something? Why not just count the number of flops and call 1000 flops a credit?[/quote]

Yes you are, the optimization of the apps, if my pc has 1000 flops and your pc has 1000 flops one would think we should get the same credits BUT since some apps are optimized and some aren't then you are getting 100 credits per day at Einstein while I'm getting 250 credits per day at MilkyWay. You crunch thru the tasks faster so you get a penalty.
ID: 76015 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mr P Hucker
Avatar

Send message
Joined: 5 Jul 11
Posts: 990
Credit: 376,143,149
RAC: 5
Message 76019 - Posted: 28 Jun 2023, 11:50:00 UTC - in response to Message 76014.  

But new and some old people don't see it that way, they see a credit as a credit as a credit. Kinda like electricity you have volts and watts and we in the US have volts and watts, you have mm's, cm's and meters while we have inches, feet and yards but they are comparable.
Those are precise measures of the quantity of the same thing. But you wouldn't compare volts to amps. My battery has 200Ah. Your battery is 24V. Which holds more power?
The above was double spaced between sentences, I apologise for the forum software ruining my post.
ID: 76019 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mr P Hucker
Avatar

Send message
Joined: 5 Jul 11
Posts: 990
Credit: 376,143,149
RAC: 5
Message 76020 - Posted: 28 Jun 2023, 11:51:15 UTC - in response to Message 76015.  

Yes you are, the optimization of the apps, if my pc has 1000 flops and your pc has 1000 flops one would think we should get the same credits BUT since some apps are optimized and some aren't then you are getting 100 credits per day at Einstein while I'm getting 250 credits per day at MilkyWay. You crunch thru the tasks faster so you get a penalty.
But a task is say 50,000 flops. If you're optimized, you do it faster, you do 50,000 in less time, so you get more credits for doing more flops today.
The above was double spaced between sentences, I apologise for the forum software ruining my post.
ID: 76020 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Link
Avatar

Send message
Joined: 19 Jul 10
Posts: 589
Credit: 18,930,552
RAC: 4,413
Message 76038 - Posted: 28 Jun 2023, 15:26:57 UTC - in response to Message 76020.  

Yes you are, the optimization of the apps, if my pc has 1000 flops and your pc has 1000 flops one would think we should get the same credits BUT since some apps are optimized and some aren't then you are getting 100 credits per day at Einstein while I'm getting 250 credits per day at MilkyWay. You crunch thru the tasks faster so you get a penalty.
But a task is say 50,000 flops. If you're optimized, you do it faster, you do 50,000 in less time, so you get more credits for doing more flops today.
That way it works only on projects, which give fixed amount of credits per task, like Einstein for example, if you use there an optimized app, you get more credits per day, that's the way it should be. It also worked like that at SETI as long as they had fixed credits according to AR.

With CreditNew, if you compute faster, it assumes simply that the task contained less work (the FLOPs estimate might be way off and has nothing to do with the credit assigned for the task, it's used just for work requests/cache management) and you end up with same credit per day as with stock app (well, not really as there are wingmen and/or similar tasks, so CreditNew will raise your average processing rate a bit and the winman's down a bit so they end up with less credits per day and you with a bit more, where it stops depends on how many use optimized apps).

The issue is that BOINC has no real idea how fast modern computers are, the benchmark tests just small fraction of the operations a modern CPU can do, MMX, all SSE, AVX and whatever else are not tested at all and project apps use them, even stock apps.

If I check some of my old computers on SETI and compare for example my AthlonXP 2000+ it with my Core 2 Duo, I see according to the benchmark values, that the C2D is about 2.5x faster on integer and 1.9x faster on floating point calculations. In reality, if I compare Astropulse performance, one core of the C2D is around 6x faster than the AthlonXP (both using optimized applications). CreditNew however sees just the benchmark values for the computers and the time the task needed and would simply assume the tasks the C2D got contained not even half the amount of work that those for the AthlonXP contained, so it will give less credits for them.
ID: 76038 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 8 May 09
Posts: 3319
Credit: 520,377,705
RAC: 23,403
Message 76077 - Posted: 28 Jun 2023, 23:32:16 UTC - in response to Message 76020.  

Yes you are, the optimization of the apps, if my pc has 1000 flops and your pc has 1000 flops one would think we should get the same credits BUT since some apps are optimized and some aren't then you are getting 100 credits per day at Einstein while I'm getting 250 credits per day at MilkyWay. You crunch thru the tasks faster so you get a penalty.


But a task is say 50,000 flops. If you're optimized, you do it faster, you do 50,000 in less time, so you get more credits for doing more flops today.


NO it doesn't work that way...you only used 40,000 flops because you used an optimzed app so you get fewer credits then I do because I did not use the optimized app.
ID: 76077 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mr P Hucker
Avatar

Send message
Joined: 5 Jul 11
Posts: 990
Credit: 376,143,149
RAC: 5
Message 76079 - Posted: 29 Jun 2023, 2:23:08 UTC

Credits aren't that important anyway, they're a tool for the project to make people run one task more than another. Primegrid for example gives more credits for their longer tasks to encourage people to do them. They're also a good way to compete with others in the same project. But I can't see a way of making it comparable across projects, fart oo many variables.
The above was double spaced between sentences, I apologise for the forum software ruining my post.
ID: 76079 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ejmwolf

Send message
Joined: 19 Jun 21
Posts: 3
Credit: 8,764,807
RAC: 0
Message 76082 - Posted: 29 Jun 2023, 6:22:47 UTC - in response to Message 75976.  

Thank you, Bill. I went to Preferences and the MilkyWay N0Body box was already checked. I unchecked the GPU boxes (not sure if that was necessary, or will affect other tasks). I'll give it a couple of days and see if I start receiving tasks.

John
ID: 76082 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mr P Hucker
Avatar

Send message
Joined: 5 Jul 11
Posts: 990
Credit: 376,143,149
RAC: 5
Message 76083 - Posted: 29 Jun 2023, 6:39:11 UTC - in response to Message 76082.  
Last modified: 29 Jun 2023, 6:39:58 UTC

If your computer has nothing to do, it should have asked. Try manually asking for work, then looking in "tools, event log" for Boinc Manager, or the messages tab in Boinctasks.

You should get something like:

Requesting new tasks for CPU	
Scheduler request completed: got 0 new tasks	
No tasks sent	
No tasks are available for Nbody
If you were to get the above saying "no tasks available", then you'll know why. You might also see "CPU cache full" if your computer already has enough work.
The above was double spaced between sentences, I apologise for the forum software ruining my post.
ID: 76083 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 8 May 09
Posts: 3319
Credit: 520,377,705
RAC: 23,403
Message 76088 - Posted: 29 Jun 2023, 12:31:38 UTC - in response to Message 76079.  

Credits aren't that important anyway, they're a tool for the project to make people run one task more than another. Primegrid for example gives more credits for their longer tasks to encourage people to do them. They're also a good way to compete with others in the same project. But I can't see a way of making it comparable across projects, fart oo many variables.


It's waaay too late to unring that bell, credits are here to stay. Credits alos help a project attract users ie Collatz used to be the best paying gpu project BY FAR and as a result had ALOT of people crunching for it, the other projects complained but the Admin was trying to prove a point so the credits stayed where they were. Other projects in the past were getting paid for the work that the project was doing, they said so up front, so they also raised the credits a bit to attract people so they could meet their deadlines. Some projects just don't care how long it takes to get the data crunched so the credits are lower, lower credits also tend to keep the hackers away that can cause database problems if you don't have enough security on your tasks. Collatz itself was hacked by someone who wanted more GridCoins then anyone else and it took almost a month for the Admin to put in the necessary security checks before they could safely send out tasks again. Collatz was a one man operation out of his home so it took a little longer than a project that's bigger and has more people to do the work.
ID: 76088 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
kotenok2000
Avatar

Send message
Joined: 22 May 11
Posts: 69
Credit: 5,658,788
RAC: 1,157
Message 76089 - Posted: 29 Jun 2023, 12:35:09 UTC

Is there a way to make boinc think cpu has more cores?
For example 2 cores when cpu has one?
ID: 76089 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mr P Hucker
Avatar

Send message
Joined: 5 Jul 11
Posts: 990
Credit: 376,143,149
RAC: 5
Message 76090 - Posted: 29 Jun 2023, 13:22:00 UTC - in response to Message 76089.  
Last modified: 29 Jun 2023, 13:23:32 UTC

Is there a way to make boinc think cpu has more cores?
For example 2 cores when cpu has one?
Probably not, but you can use an app config to tell a project each task uses 0.5 cores:

<app_config>
    <app_version>
        <app_name>insertname</app_name>
        <plan_class></plan_class>
        <avg_ncpus>0.5</avg_ncpus>
    </app_version>
</app_config>

The above was double spaced between sentences, I apologise for the forum software ruining my post.
ID: 76090 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Link
Avatar

Send message
Joined: 19 Jul 10
Posts: 589
Credit: 18,930,552
RAC: 4,413
Message 76092 - Posted: 29 Jun 2023, 14:00:12 UTC - in response to Message 76089.  

Is there a way to make boinc think cpu has more cores?
For example 2 cores when cpu has one?
<cc_config>
 <options>
  <ncpus>2</ncpus>
 </options>
</cc_config>

ID: 76092 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Previous · 1 . . . 8 · 9 · 10 · 11 · 12 · 13 · 14 . . . 16 · Next

Message boards : News : Separation Application Shutting Down on Tuesday, Jun 20th

©2024 Astroinformatics Group