2. For this question, what I'm asking is, if I have 4 cores (which I do) and I lie and say I have 6 will it open and run 6 tasks? What are the ramifications?

You can lie and tell Boinc you have any arbitrary number of CPUs using the <ncpus> tag in the cc_config.xml ovverride file. It has sense for testing purposes, but will cause unnecessary context switches and thrashing the CPU cache (because more applications will share less cores).

I was personally doing this while crunching Malaria WUs - they were (or still are?) prone to stop crunching (or not being preempted, or not starting to consume CPU time after being selected) and effectively causing the CPU to sit idle until someone (me) will notice it. Forcing Boinc scheduler to behave as with 2 CPUs caused to run apps much less effectively, but in case of such acident the other app could run using he whole CPU.

Peter

I appreciate your perspective on this Peter. Thanks for responding.

---

Ty < finally.. thinks he's got it fixed now

2. For this question, what I'm asking is, if I have 4 cores (which I do) and I lie and say I have 6 will it open and run 6 tasks? What are the ramifications?

You can lie and tell Boinc you have any arbitrary number of CPUs using the <ncpus> tag in the cc_config.xml ovverride file. It has sense for testing purposes, but will cause unnecessary context switches and thrashing the CPU cache (because more applications will share less cores).

I was personally doing this while crunching Malaria WUs - they were (or still are?) prone to stop crunching (or not being preempted, or not starting to consume CPU time after being selected) and effectively causing the CPU to sit idle until someone (me) will notice it. Forcing Boinc scheduler to behave as with 2 CPUs caused to run apps much less effectively, but in case of such acident the other app could run using he whole CPU.

Peter

I appreciate your perspective on this Peter. Thanks for responding.

I decided to take a look at this config file. Search the entire C: drive. Can't find one. Is this part of a source release or something only available in a Linux environment, or is it an optional override file I don't have?

On my question number 3 regarding limiting active tasks per project. I discovered I can sort of get the behavior I want by setting the parameter that controls days of work in advance to zero. What sort of development environment do I need to change the BOINC source to do what I want myself? I hate brick walls.

---

Ty < finally.. thinks he's got it fixed now

I decided to take a look at this config file. Search the entire C: drive. Can't find one. Is this part of a source release or something only available in a Linux environment, or is it an optional override file I don't have?

The core client configuration file is something you have to make your self.
See How do I configure my client using the cc_config.xml file? for more information.

What sort of development environment do I need to change the BOINC source to do what I want myself? I hate brick walls.

Software prerequisites (Unix/Linux) and the source code.

---

I decided to take a look at this config file. Search the entire C: drive. Can't find one. Is this part of a source release or something only available in a Linux environment, or is it an optional override file I don't have?

The core client configuration file is something you have to make your self.
See How do I configure my client using the cc_config.xml file? for more information.

What sort of development environment do I need to change the BOINC source to do what I want myself? I hate brick walls.

Software prerequisites (Unix/Linux) and the source code.

Reply:

I'm in a catch 22 on putting up a Unix/Linux environment because I have to transition from XP to Win Server 2003 X64 in order to host Unix as a guest OP sys on the server. MS does not recommend trying to do guesting on XP. I could set up to boot it (did that once) but that's another can of worms so I guess I won't be developing on BOINC in the near term. I can take a look at the source code and may do that a little bit. The htm I can build right away.

BTW in the last day of running on this new release I picked up 1 teeny bug. The HELP/BOINC website pull down opens the BOINC manager help instead of the main BOINC page. Used to work right. Small nit.

Thanks again.

---

Ty < finally.. thinks he's got it fixed now

You can download the source code with Tortoise SVN. The link to use in there is http://boinc.berkeley.edu/svn/trunk/boinc

Compiling it under Windows will need a (usually registered) compiler though. Look at Intel compilers.

---

You can download the source code with Tortoise SVN. The link to use in there is http://boinc.berkeley.edu/svn/trunk/boinc

Compiling it under Windows will need a (usually registered) compiler though. Look at Intel compilers.

Feedback

So far I've had the opportunity to build a cc_config.xml and turn on a few things. Right off I found out that Einstein checkpoints every couple of seconds and ignores the disk write settings. Wow! I also opened my processor up to six cpu's (lied). Actually works quite well most likely because of the capacity in this box. I believe everything has room to just sit in core. No significant thrashing to page file or disk observed. I also notice that no single project can use more than 1 cpu on my box no matter how many open tasks it has. That's a good behavior from my perspective anyway. BOINC still has no issues with downloading as many tasks as I have declared CPUs. Don't like that part at all but at least the # of tasks downloaded is no longer exceeding the CPU count.

The last parameter in the config file isn't parsing. Don't know how to make it happy. <work_request_factor>1</work_request_factor>
I'm real interested in that one obviously.

My ideal environment would be at most two open tasks per project (hard limit), nearly all in running state and no backlog whatsoever. If a project has no work I don't care. BOINC scheduling appears to be deadline oriented. I don't really care about that either. Maybe someday after I've been using BOINC for awhile I'll learn to appreciate the feature. :)

I downloaded and installed Tortoise. "Import" command required a password. Maybe that's the wrong command to use. I started browsing. Can do that! I'll take a hard look at the code and see if I can locate the right place for an "opportunity" in the cracks around my other work. I primarily code in COBOL for a living but I can work in C as well. I'm just not nearly as experienced in it and I'm real busy so it will be slow going. I'll get back to you. Thank you for your responses.

Ty

---

Ty < finally.. thinks he's got it fixed now

I downloaded and installed Tortoise. "Import" command required a password. Maybe that's the wrong command to use. I started browsing. Can do that! I'll take a hard look at the code and see if I can locate the right place for an "opportunity" in the cracks around my other work. I primarily code in COBOL for a living but I can work in C as well. I'm just not nearly as experienced in it and I'm real busy so it will be slow going. I'll get back to you. Thank you for your responses.

Ty

You'll want to right click on a new directory (I called mine BOINC Source) and choose SVN Update, not import from that popout menu. Then put in the URL.

I just created a clean directory on a different partition and it works fine.

Assuming Windows of course...

---

Kathryn :o)

[/quote]

The last parameter in the config file isn't parsing. Don't know how to make it happy. <work_request_factor>1</work_request_factor>
I'm real interested in that one obviously.

Ty

[/quote]

I think that parameter was supplanted by: Maintain enough work for an additional
(Requires 5.10+ client under Network in General Preferences.

It has been working for me since I updated to 5.10.8.

---

I downloaded and installed Tortoise. "Import" command required a password. Maybe that's the wrong command to use. I started browsing. Can do that! I'll take a hard look at the code and see if I can locate the right place for an "opportunity" in the cracks around my other work. I primarily code in COBOL for a living but I can work in C as well. I'm just not nearly as experienced in it and I'm real busy so it will be slow going. I'll get back to you. Thank you for your responses.

Ty

You'll want to right click on a new directory (I called mine BOINC Source) and choose SVN Update, not import from that popout menu. Then put in the URL.

I just created a clean directory on a different partition and it works fine.

Assuming Windows of course...

Thanks Kathryn
I'll give it a try in a bit. BTW we celebrate Eore's Birthday in here in Austin every year.

---

Ty < finally.. thinks he's got it fixed now

The last parameter in the config file isn't parsing. Don't know how to make it happy. <work_request_factor>1</work_request_factor>
I'm real interested in that one obviously.

Ty

[/quote]

I think that parameter was supplanted by: Maintain enough work for an additional
(Requires 5.10+ client under Network in General Preferences.

It has been working for me since I updated to 5.10.8.
[/quote]

I installed 5.10.7 several days ago. I hope it wasn't supplanted. I have have "maintain" set to zero. I don't want any additional work. Thanks for responding.

---

Ty < finally.. thinks he's got it fixed now

The last parameter in the config file isn't parsing. Don't know how to make it happy. <work_request_factor>1</work_request_factor>
I'm real interested in that one obviously.

Ty

I think that parameter was supplanted by: Maintain enough work for an additional
(Requires 5.10+ client under Network in General Preferences.

It has been working for me since I updated to 5.10.8.

I installed 5.10.7 several days ago. I hope it wasn't supplanted. I have have "maintain" set to zero. I don't want any additional work. Thanks for responding.

But that's exactly what that flag did. It allowed extra caching of work. So I'm a bit confused why you want that rather than the current web/local version.

---

Kathryn :o)

I downloaded and installed Tortoise. "Import" command required a password. Maybe that's the wrong command to use. I started browsing. Can do that! I'll take a hard look at the code and see if I can locate the right place for an "opportunity" in the cracks around my other work. I primarily code in COBOL for a living but I can work in C as well. I'm just not nearly as experienced in it and I'm real busy so it will be slow going. I'll get back to you. Thank you for your responses.

Ty

You'll want to right click on a new directory (I called mine BOINC Source) and choose SVN Update, not import from that popout menu. Then put in the URL.

I just created a clean directory on a different partition and it works fine.

Assuming Windows of course...

Thanks Kathryn
I'll give it a try in a bit. BTW we celebrate Eore's Birthday in here in Austin every year.

Let me know if it doesn't work for you.


I lived in Austin for several years while I was in grad school. I think I went one year while I was there.

---

Kathryn :o)

The last parameter in the config file isn't parsing. Don't know how to make it happy. <work_request_factor>1</work_request_factor>
I'm real interested in that one obviously.

Ty

I think that parameter was supplanted by: Maintain enough work for an additional
(Requires 5.10+ client under Network in General Preferences.

It has been working for me since I updated to 5.10.8.
[/quote]

I installed 5.10.7 several days ago. I hope it wasn't supplanted. I have have "maintain" set to zero. I don't want any additional work. Thanks for responding.[/quote]

You could just delete the <work_request_factor (entire line) from the config file then.

---

snipped...

But that's exactly what that flag did. It allowed extra caching of work. So I'm a bit confused why you want that rather than the current web/local version.

If you set the 'connect to network' setting to 0.001, the results are reported almost immediately, which is what quite a few users have been asking for (why?), for some time. And then use the 'work request factor' to set your cache.

Andy

snipped...

But that's exactly what that flag did. It allowed extra caching of work. So I'm a bit confused why you want that rather than the current web/local version.

If you set the 'connect to network' setting to 0.001, the results are reported almost immediately, which is what quite a few users have been asking for (why?), for some time. And then use the 'work request factor' to set your cache.
Andy

That's not very efficient for any of the project servers!!!!!!!

I realise that, the reason I included the (why?).

In fact since that post I'm wondering wether the code that normally delays reports until there is a request for work, should be reversed. So that requests for work are delayed until a report is triggered, unless the host is liable to run out of work.

Andy

If all is well, JM7 did report this to David. I don't see a change in the code yet, though.

Edit: I just spoke with David.
JM7 sent in a fix that will be included in the next release.

---

"Wow!!" I need to watch this thread a little closer. Haven't checked on it in a few hours and look at all the commotion I caused over that flag.

I think is worth revisiting my objective again. Like I said in an earlier posting "My ideal environment would be at most two open tasks per project (hard limit), nearly all in running state and no backlog whatsoever. If a project has no work I don't care."

There is a good reason for this. When I allow my computer to take on work from one of these projects I feel some obligation to the project to do whatever I can to get the results done and back in a timely fashion. Also I use this computer for software development and sometimes I have to take it off line to make changes to hardware or the OS. For these reasons I'd rather have the computer be partly idle than risk accidently trashing a bunch of unfinished work. I can take on about two concurrently running tasks during the day and six or even more over night. They have to be very short or good at recovery. I don't want to have a backlog of tasks waiting to run. I'd rather they be loaded and cranking or not there at all.

Adding more CPU threads gets me part way there. Shutting off the ability of BOINC to take on more work than I can have concurrently loaded and executing gets me further. BOINC can still run unbalanced though in this scenario. One project can download as many tasks as I have CPU's and shut the others out for days. I've been using the "no new tasks" button to keep the mix balanced but if there is a flag anywhere that will let me say "only 1 or only 2 tasks per project then I want to set that flag. I think its trying to balance the hours contributed or something. An algorithm like that will favor a newly added project over the others and it could take quite a while for everything to balance.

I know this is volunteer oriented here so eventually I may get around to making the changes myself but my personal work queue is very high too. So for now I'll just go with what's available and manually tweek it now and then.

I really appreciate everyone's interest and attention this. It's way more than I expected. Really!

You're all very kind.

---

Ty < finally.. thinks he's got it fixed now

The last parameter in the config file isn't parsing. Don't know how to make it happy. <work_request_factor>1</work_request_factor>
I'm real interested in that one obviously.

Ty

I think that parameter was supplanted by: Maintain enough work for an additional
(Requires 5.10+ client under Network in General Preferences.

It has been working for me since I updated to 5.10.8.

I installed 5.10.7 several days ago. I hope it wasn't supplanted. I have have "maintain" set to zero. I don't want any additional work. Thanks for responding.

But that's exactly what that flag did. It allowed extra caching of work. So I'm a bit confused why you want that rather than the current web/local version.

I don't know what the default is so setting it for minimum takes it out of play.

---

Ty < finally.. thinks he's got it fixed now

I downloaded and installed Tortoise. "Import" command required a password. Maybe that's the wrong command to use. I started browsing. Can do that! I'll take a hard look at the code and see if I can locate the right place for an "opportunity" in the cracks around my other work. I primarily code in COBOL for a living but I can work in C as well. I'm just not nearly as experienced in it and I'm real busy so it will be slow going. I'll get back to you. Thank you for your responses.

Ty

You'll want to right click on a new directory (I called mine BOINC Source) and choose SVN Update, not import from that popout menu. Then put in the URL.

I just created a clean directory on a different partition and it works fine.

Assuming Windows of course...

Thanks Kathryn
I'll give it a try in a bit. BTW we celebrate Eore's Birthday in here in Austin every year.

Let me know if it doesn't work for you.


I lived in Austin for several years while I was in grad school. I think I went one year while I was there.

I got really hacked at BOINC manager tonight. I went to manually allow some new tasks to download and it freed up the wrong project. Started downloading multiple CPDN tasks and one other. I aborted them immediately. I want that one off this machine. I have server I can run it on. So I tried again to download the source to fix this. There is not a SVN Update on the tortoise version I have but there is an SVN Checkout. Now, I didn't want to checkout any source to change, just wanted to get a copy of it. But there were no other choices to run so I selected checkout. It downloaded but it didn't end right. Errored off. After looking at the code for over an hour I decided I really needed to look at cs_scheduler.C or something close to that and it wasn't there. I give up for tonight. I'll try again tomorrow.

---

Ty < finally.. thinks he's got it fixed now

I want to upgrade to version 5.10.13, however, I don't see any specific guidance about upgrading anywhere.

Do I just install over my existing version (5.10.7) or do I have to uninstall it first?

Thanks,
Ed