Welcome to MilkyWay@home

New App status

Message boards : Number crunching : New App status
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 . . . 10 · Next

AuthorMessage
Profile banditwolf
Avatar

Send message
Joined: 12 Nov 07
Posts: 2425
Credit: 524,164
RAC: 0
Message 5307 - Posted: 6 Oct 2008, 15:01:19 UTC - in response to Message 5300.  


What is the status of the new application - when can we expect it to be released ?


I would appreciate a response, as I assume this is about an optimised Milkyway client?


I don't think that is the intent. It is supposed to be a whole new code. Optimizing doesn't seem to be the focus from what I understand. The users are supposed to submit their optimized code to check for validity.


I know Travis is around since he has posted in the 'Application Code Discussion' thread. And yet this goes on without a response. It must be asking too much to take 1 minute to type a quick reply.
Doesn't expecting the unexpected make the unexpected the expected?
If it makes sense, DON'T do it.
ID: 5307 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Dave Przybylo
Avatar

Send message
Joined: 5 Feb 08
Posts: 236
Credit: 49,648
RAC: 0
Message 5309 - Posted: 6 Oct 2008, 17:38:04 UTC - in response to Message 5307.  


What is the status of the new application - when can we expect it to be released ?


I would appreciate a response, as I assume this is about an optimised Milkyway client?


I don't think that is the intent. It is supposed to be a whole new code. Optimizing doesn't seem to be the focus from what I understand. The users are supposed to submit their optimized code to check for validity.


I know Travis is around since he has posted in the 'Application Code Discussion' thread. And yet this goes on without a response. It must be asking too much to take 1 minute to type a quick reply.



It's just that we've said over and over that this isn't about an optimized application and people still keep asking about it. That isn't our intent. Our intent is to get the science correct.
Dave Przybylo
MilkyWay@home Developer
Department of Computer Science
Rensselaer Polytechnic Institute
ID: 5309 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Milksop at try

Send message
Joined: 1 Oct 08
Posts: 106
Credit: 24,162,445
RAC: 0
Message 5311 - Posted: 6 Oct 2008, 18:22:15 UTC - in response to Message 5309.  
Last modified: 6 Oct 2008, 18:26:33 UTC

It's just that we've said over and over that this isn't about an optimized application and people still keep asking about it. That isn't our intent. Our intent is to get the science correct.

Maybe you have done the second step before the first?
You could have run the project on a small scale inside your institute. When you have something halfway appropriate you could still have opened a Boinc project.

But in the current state the waste of resources by MilkyWay@home is just lunatic. I think no responsible cruncher can run MW anymore, if he knows about the extent of this waste.
ID: 5311 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Dave Przybylo
Avatar

Send message
Joined: 5 Feb 08
Posts: 236
Credit: 49,648
RAC: 0
Message 5312 - Posted: 6 Oct 2008, 18:37:54 UTC - in response to Message 5311.  

It's just that we've said over and over that this isn't about an optimized application and people still keep asking about it. That isn't our intent. Our intent is to get the science correct.

Maybe you have done the second step before the first?
You could have run the project on a small scale inside your institute. When you have something halfway appropriate you could still have opened a Boinc project.

But in the current state the waste of resources by MilkyWay@home is just lunatic. I think no responsible cruncher can run MW anymore, if he knows about the extent of this waste.


What's the waste? We still have calculated many many stars. We're just looking to get the results to become more precise.
Dave Przybylo
MilkyWay@home Developer
Department of Computer Science
Rensselaer Polytechnic Institute
ID: 5312 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Milksop at try

Send message
Joined: 1 Oct 08
Posts: 106
Credit: 24,162,445
RAC: 0
Message 5314 - Posted: 6 Oct 2008, 18:53:06 UTC - in response to Message 5312.  

It's just that we've said over and over that this isn't about an optimized application and people still keep asking about it. That isn't our intent. Our intent is to get the science correct.

Maybe you have done the second step before the first?
You could have run the project on a small scale inside your institute. When you have something halfway appropriate you could still have opened a Boinc project.

But in the current state the waste of resources by MilkyWay@home is just lunatic. I think no responsible cruncher can run MW anymore, if he knows about the extent of this waste.


What's the waste? We still have calculated many many stars. We're just looking to get the results to become more precise.

Just look at my profile and you will get a glimpse of the wasted resources.
The crunchers out there could very well calculate the same results as now in two percent of the time they spend now.
The time it takes to calculated a Milkyway WU could very well be better spend to run other projects until you get your things together.

And don't hide behinde the sentence: "we want the results to be more precise before we optimize". If I have seen it right in the code discussion thread there was a serious bug in the code impeding your "precision". You should have found such things yourself, if precision is your highest goal.

And frankly, when looking at the source code one can get some doubts if the MW staff is actually able to put some reasonable code together. You should look for a fresh and ambitious undergraduate or PhD student!
ID: 5314 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Opteron

Send message
Joined: 1 Oct 08
Posts: 5
Credit: 1,892
RAC: 0
Message 5315 - Posted: 6 Oct 2008, 18:56:35 UTC - in response to Message 5312.  

What's the waste? We still have calculated many many stars. We're just looking to get the results to become more precise.


Ehh interesting, if I interpret Milksop's profile correct, you could have calculated ~50 times more stars so far.

Would that be bad ?
ID: 5315 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Fab36-Folding-Division

Send message
Joined: 3 Sep 07
Posts: 1
Credit: 10,398
RAC: 0
Message 5316 - Posted: 6 Oct 2008, 19:10:16 UTC - in response to Message 5312.  

What's the waste?
It's the waste of energy - the energy paid by your supporters! You're right that in this stage of MW correct results should count more than performance. But this doesn't mean you should write code that takes hours for something that can easily (i.e. without doing rocket science) be done within minutes. E.g. I've seen code like this in the MW sources:
x = y / z;
v = x * z;
Anyone who made it through school's Mathmatics, will notice that this is the same as v = y. Actually the redundant division decreases accuracy and therefore it should be in YOUR interest to change this code. You may say it's just a division and a multiplication. But a division isn't a simple operation and the code was called very often.

I think noone expects you to deliver a highly-optimized app as long as MW is in Alpha stage. But what we have now isn't a "normal" app, it's a highly de-optimized app and that's not good enough even for Alpha stage.
ID: 5316 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Milksop at try

Send message
Joined: 1 Oct 08
Posts: 106
Credit: 24,162,445
RAC: 0
Message 5317 - Posted: 6 Oct 2008, 19:15:57 UTC - in response to Message 5316.  

You're right that in this stage of MW correct results should count more than performance. But this doesn't mean you should write code that takes hours for something that can easily (i.e. without doing rocket science) be done within minutes. E.g. I've seen code like this in the MW sources:
x = y / z;
v = x * z;
Anyone who made it through school's Mathmatics, will notice that this is the same as v = y. Actually the redundant division decreases accuracy and therefore it should be in YOUR interest to change this code. You may say it's just a division and a multiplication. But a division isn't a simple operation and the code was called very often.

I think noone expects you to deliver a highly-optimized app as long as MW is in Alpha stage. But what we have now isn't a "normal" app, it's a highly de-optimized app and that's not good enough even for Alpha stage.

I have put that example even in my profile, because it is that absurd ;)
ID: 5317 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile banditwolf
Avatar

Send message
Joined: 12 Nov 07
Posts: 2425
Credit: 524,164
RAC: 0
Message 5318 - Posted: 6 Oct 2008, 19:23:40 UTC

at Dave: Is the new app. actually being worked on? It has been a LOT longer than initially said.
Doesn't expecting the unexpected make the unexpected the expected?
If it makes sense, DON'T do it.
ID: 5318 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Dave Przybylo
Avatar

Send message
Joined: 5 Feb 08
Posts: 236
Credit: 49,648
RAC: 0
Message 5320 - Posted: 6 Oct 2008, 19:36:04 UTC - in response to Message 5318.  
Last modified: 6 Oct 2008, 19:36:39 UTC

at Dave: Is the new app. actually being worked on? It has been a LOT longer than initially said.


It is and we're working what was just said in. We're only going to be doing that calculation once. This is good progress. Now, if the people suggesting that we should get our act together could actually post in the application code suggestion board with code changes that would actually be proactive it would get things done much faster. We did not write most of this code. Astronomy students wrote this because it is essentially their project. We were just given the code to "BOINCify" it. We now have the problem of streamlining it and making it better and any help would be much appreciated. I can see some people have already done such things and my question is why would you hold it back from the community? Make code suggestions! This project is open to the community for the betterment of it. If people who used open source projects just kept the changes to themselves, the project would never go anywhere. We need people to suggest improvements in the code and help us out too.
Dave Przybylo
MilkyWay@home Developer
Department of Computer Science
Rensselaer Polytechnic Institute
ID: 5320 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Milksop at try

Send message
Joined: 1 Oct 08
Posts: 106
Credit: 24,162,445
RAC: 0
Message 5321 - Posted: 6 Oct 2008, 20:03:37 UTC - in response to Message 5320.  
Last modified: 6 Oct 2008, 20:15:52 UTC

Now, if the people suggesting that we should get our act together could actually post in the application code suggestion board with code changes that would actually be proactive it would get things done much faster. We did not write most of this code. Astronomy students wrote this because it is essentially their project. We were just given the code to "BOINCify" it. We now have the problem of streamlining it and making it better and any help would be much appreciated. I can see some people have already done such things and my question is why would you hold it back from the community? Make code suggestions! This project is open to the community for the betterment of it. If people who used open source projects just kept the changes to themselves, the project would never go anywhere. We need people to suggest improvements in the code and help us out too.

Just look at the code!
Everyone with some interest and capabilities in C programming should get a factor of 10 out of it. Personally, I haven't done some coding for some years before looking at MW and that speedup was just a matter of hours (with me having absolutly no experience in compiling a BOINC project). Half of the time was simply needed to grasp what is actually done in what function and the ways the program calls all these funtions.

The most important thing is of course the innermost loop that is done in the qgaus function and the other functions that are called from there. That was said to you already in the code discussion thread. And I would bet Crunch3r told you similiar things already half a year ago.

And look what is really needed in the innermost loop! Extremely often there are constructs like this:
for(i=0;i<imax;i++)
{
  for(j=0;j<jmax;j++)
  {
    abc=func(i);
    xyz=func2(abc*i);
    result+=func3(xyz+j);
  }
}
Some things are really calculated several hundred million times in the innermost loop, that simply don't change for any given WU. Try to avoid divisions and pow() and exp() calls, as these are very costly in terms of CPU cycles. But all this is common sense and you should not need the advice by some users to get such things done.

And the golden rule for MW should be: The fastest function is the function that does not get called.
ID: 5321 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Milksop at try

Send message
Joined: 1 Oct 08
Posts: 106
Credit: 24,162,445
RAC: 0
Message 5322 - Posted: 6 Oct 2008, 20:07:40 UTC - in response to Message 5320.  
Last modified: 6 Oct 2008, 20:08:31 UTC

Just an addon:
my question is why would you hold it back from the community?

Do you want us to release the binary? You should better be prepared for that!
ID: 5322 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Dave Przybylo
Avatar

Send message
Joined: 5 Feb 08
Posts: 236
Credit: 49,648
RAC: 0
Message 5325 - Posted: 6 Oct 2008, 21:06:20 UTC - in response to Message 5322.  

Just an addon:
my question is why would you hold it back from the community?

Do you want us to release the binary? You should better be prepared for that!


We want you to release the code to the community that you've updated. I mean there are thousands of you out there while there are only a very few of us in here.
Dave Przybylo
MilkyWay@home Developer
Department of Computer Science
Rensselaer Polytechnic Institute
ID: 5325 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Milksop at try

Send message
Joined: 1 Oct 08
Posts: 106
Credit: 24,162,445
RAC: 0
Message 5326 - Posted: 6 Oct 2008, 21:41:28 UTC - in response to Message 5325.  

We want you to release the code to the community that you've updated. I mean there are thousands of you out there while there are only a very few of us in here.

Crunch3r was alone some months ago, I was alone and my partner at this account (he has created an own version of the app) was also alone when looking at the source. We all three did not share the code with each other. Nevertheless we all arrived at comparable speedups far greater than factor 10 (more like 30 to 70) in quite short time. And of course, we also did not write the original code ourselves. Does this say something about the code or the project?

Yes, I think!
The MW project appears to be unable or unwilling to come up with some very basic improvements to the code. Therefore one has to have some doubts, if they are able to do the things they are actually required to do. It is not a solution if some volunteer basically writes up a new app and nobody is able to change something on it afterwards.

Probably it is the way Travis just wrote on the homepage and you here in the thread, that some astronomy students have put together a horrible app. They were of course no computer scientists. But if I'm not completly mistaken, you, Travis and Nate are supposed to be these computer scientists, isn't? I don't know the profession of Crunch3r and also not that of my account partner. But I can tell you that I don't earn my money with writing code nor I ever have or plan to do so.

You should draw your conclusions. My suggestion would be that you three (maybe there are more, I missed) just sit down and look at the code. When you come up with something reasonable (should be possible in quite short time). We will step down and help. Look at our profile for the small challenge. It is serious.
ID: 5326 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Brian Silvers

Send message
Joined: 21 Aug 08
Posts: 625
Credit: 558,425
RAC: 0
Message 5329 - Posted: 7 Oct 2008, 0:02:50 UTC - in response to Message 5326.  


You should draw your conclusions. My suggestion would be that you three (maybe there are more, I missed) just sit down and look at the code. When you come up with something reasonable (should be possible in quite short time). We will step down and help. Look at our profile for the small challenge. It is serious.


I disliked how Travis treated Matthias and I stated that here publicly, but two wrongs do not make a right. I would encourage you to stop trying to "take Travis and Dave to school". They have asked for your help. If you see something, mention it. They have even gone to the trouble to set up a place for you all to talk about coding issues with the app. They are telling you that their job is not to make it go fast, but to try to make it accurate. If you are able to help do both, it certainly appears as though they are willing to listen at this point.

Why don't you try working together rather than to try to prove who was right and who was wrong with the whole incident a few months ago?
ID: 5329 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Misfit
Avatar

Send message
Joined: 27 Aug 07
Posts: 915
Credit: 1,503,319
RAC: 0
Message 5331 - Posted: 7 Oct 2008, 0:25:15 UTC - in response to Message 5321.  

And the golden rule for MW should be: The fastest function is the function that does not get called.

That sounds more like a REM statement.
me@rescam.org
ID: 5331 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Brian Silvers

Send message
Joined: 21 Aug 08
Posts: 625
Credit: 558,425
RAC: 0
Message 5332 - Posted: 7 Oct 2008, 1:39:09 UTC - in response to Message 5331.  

And the golden rule for MW should be: The fastest function is the function that does not get called.

That sounds more like a REM statement.


Leave it to you to make a comment like that...
ID: 5332 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
stwainer

Send message
Joined: 20 Nov 07
Posts: 4
Credit: 188,274
RAC: 0
Message 5347 - Posted: 7 Oct 2008, 22:45:25 UTC - in response to Message 5332.  

As an ancient computer programmer, that bit o' code that Milksop put up looks a lot like some poor person converted an old FORTRAN 77 program into C code without really any understanding of what he/she was doing. Stuff like that happens all the time.

Since I'm ancient, and don't code in C, I can't help very much but would like those people who are to help these guys out.

It's easy to complain and assign blame and certainly some of it is deserved, but volunteer computing is about volunteering, right?

Dave, the moderator, made what I thought was a good point. There's only a few of them behind the curtain and thousands of us out here. Chances are that out here someone could help them with the application. From reading this thread it seems that there are people (at least two) who could make a few, simple for them, changes and things would be better quickly.

Why don't we do that? What's the hold up? Are peoples egos so bruised from the name calling that we just can't get on with things?


ID: 5347 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Milksop at try

Send message
Joined: 1 Oct 08
Posts: 106
Credit: 24,162,445
RAC: 0
Message 5348 - Posted: 8 Oct 2008, 0:10:31 UTC - in response to Message 5331.  
Last modified: 8 Oct 2008, 0:27:37 UTC

And the golden rule for MW should be: The fastest function is the function that does not get called.

That sounds more like a REM statement.

You are exactly right!
I mentioned already several times the qgaus function forming the innermost loop of the Milkyway app (it was also mentioned in the code discussion one week ago). In fact, I commented out 12 lines there (including a call to a quite time consuming function). Now only 4 lines are left, 2 of them variable declarations, the other one are the actual loop and the last one the return statement. Actually it is almost a shame, to make a function out of only one useful line of code. But at least the compiler inlines it automatically.

These are the things that really cast a doubt on the capabilities of the guys responsible for the application code. Even a half-baked "computer scientist" should be able to correct such atrocities in some minutes.
ID: 5348 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Milksop at try

Send message
Joined: 1 Oct 08
Posts: 106
Credit: 24,162,445
RAC: 0
Message 5350 - Posted: 8 Oct 2008, 1:05:17 UTC - in response to Message 5329.  
Last modified: 8 Oct 2008, 1:14:13 UTC

I disliked how Travis treated Matthias and I stated that here publicly, but two wrongs do not make a right. I would encourage you to stop trying to "take Travis and Dave to school". They have asked for your help. If you see something, mention it. They have even gone to the trouble to set up a place for you all to talk about coding issues with the app. They are telling you that their job is not to make it go fast, but to try to make it accurate. If you are able to help do both, it certainly appears as though they are willing to listen at this point.

Why don't you try working together rather than to try to prove who was right and who was wrong with the whole incident a few months ago?

I guess Matthias is Crunch3rs real name?

You have to understand that Crunch3r was a volunteer developer at this project and already improved the performance of the official app a lot. As I understood, he told the project that huge gains with little effort are possible already half a year ago. But no one of the project has taken the necessary steps. Why?

And I am actually trying to avoid "taking Dave and Travis to school". I just told them, they should do their homework. If we start to discuss the code in the forum, then we will take them to school. And you can bet, it would be embarrassing for them. So it should be really in their own interest to implement the basic improvements alone. It isn't rocket science.

Concerning the excuse with the accuracy versus speed, my first question to Dave was if they may have done the second step before the first. They could have ran all these tests completely inhouse. Is it asked too much, that they look to their code for a day or two before opening a Boinc project? For what do we pay our energy bills? Couldn't we spend our crunching power better at other projects?

Otherwise, my answer to Dave still stands:
The MW project appears to be unable or unwilling to come up with some very basic improvements to the code. Therefore one has to have some doubts, if they are able to do the things they are actually required to do. It is not a solution if some volunteer basically writes up a new app and nobody is able to change something on it afterwards.
ID: 5350 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Previous · 1 · 2 · 3 · 4 · 5 . . . 10 · Next

Message boards : Number crunching : New App status

©2024 Astroinformatics Group