BOFH = Bastard Operator From Hell
BOFH = Bastard Operator From Hell
I KNEW there was something you weren't telling us Poppa....
It seems that being forgetful can be convenient at times
Me? I'm just getting there
oh, and I found an old pic of D_A's Linux teacher from back awhile
Last edited by OldChap; 11-14-2013 at 10:30 AM.
My Biggest Fear Is When I die, My Wife Sells All My Stuff For What I Told Her I Paid For It.79 SB threads and 32 IB Threads across 4 rigs 111 threads Crunching!!
OK, back to "nice" and "priority" and things.
I put <no_priority_change>1</no_priority_change> in the cc_config.xml file. Boinc sees it in the event log
But in the bottom box is that machine. Top one is same hardware and Ubuntu version and Boinc version without the <no_priority_change>1</no_priority_change>11/14/2013 4:04:37 PM | | Config: run apps at regular priority
So Boinc sees it, but everything is still nice but lower from 19 to 10. %ni is still 99.xx and %us0.0 basically. Boinc was stopped and restarted.
So help the bearded one figure this out as I find little in searches, some lead back to here.
Why do I have a problem being "nice".
EDIT:
Since the <no_priority_change>1</no_priority_change> changes the "nice" value to that of boinc-client then I guess I need to change the "nice" value to boinc. Is that what you guys did?
OK
sudo renice 0 4107 -u boinc
did it. No smoke or flames, screams or alarms. Will watch and see.
Last edited by PoppaGeek; 11-14-2013 at 03:12 PM.
This IS changing the priority of BOINC to the same priority as the client on which boinc runs. at least that is how I see it.
I made no other changes to effect the resulting changeover from nice to the OS's sheduler
try a re-boot maybe These are running on the host here ....are yours in a vm maybe?
My Biggest Fear Is When I die, My Wife Sells All My Stuff For What I Told Her I Paid For It.79 SB threads and 32 IB Threads across 4 rigs 111 threads Crunching!!
No VM. I don't know why Boinc is running at 10 by default on mine but it is on both. I had to renice it to 0. Seems to be running ok. Reading now on how to do it on bootup but getting tired. Had several things I wanted to do here with upgrading boinc and documenting it and that is all done, unless someone finds a error or has a problem with it.
Will work on this more tomorrow. Will watch the 2 machines and see if I can tell a difference. They are both dual quad core opterons same Ubuntu and Bonic.
What I did was change the priority of the work units to the same as Boinc, the client, using the cc_config.xml then did a renice of Boinc, the client, to 0 which then set the work units to the same.
ANother thought. Difference could be processors. You are using Intel mine AMD and they do NOT do everything identically.
Last edited by PoppaGeek; 11-14-2013 at 03:33 PM.
It's "nice" because it's a system process not a "user" process. "Nice" is the priority value, and a system utility, used by the scheduler to determine which process should run at higher or lower priority and receive more or less CPU time. A process's "nice" score is relative to that of the process that called it. BOINC runs with a priority of 20 (low) and spawns worker processes with a "nice" of 19. Combined that's a priority of 39, barely above idle. This is what BOINC is designed to do. The only time nice is a problem is when the system thermal throttling acts with a higher priority.
http://en.wikipedia.org/wiki/Nice_(Unix)
If you REALLY want to mess with it, set up a cron job as root to renice the WCG processes to -20. That might require a regular expression to sort the process out.
OR alter the boinc-client start script (/etc/init.d/boinc-client from memory) and add the command nice --20 to start of the line that launches the boinc binary. I don't think messing with this a a great idea, but if you're going to do it then this might be more effective.
[SIGPIC][/SIGPIC]
I am just curious why I am getting results than Rob_B and OC.
The metric that I watch is that of cpu time vs real time in the results pages on WCG which has gone from minutes to seconds per wu which are typically running in excess of 4 hours each and often a lot longerOriginally Posted by PoppaGeek
No VM. I don't know why Boinc is running at 10 by default on mine but it is on both. I had to renice it to 0. Seems to be running ok. Reading now on how to do it on bootup but getting tired. Had several things I wanted to do here with upgrading boinc and documenting it and that is all done, unless someone finds a error or has a problem with it.
Will work on this more tomorrow. Will watch the 2 machines and see if I can tell a difference. They are both dual quad core opterons same Ubuntu and Bonic.
What I did was change the priority of the work units to the same as Boinc, the client, using the cc_config.xml then did a renice of Boinc, the client, to 0 which then set the work units to the same.
ANother thought. Difference could be processors. You are using Intel mine AMD and they do NOT do everything identically.
My Biggest Fear Is When I die, My Wife Sells All My Stuff For What I Told Her I Paid For It.79 SB threads and 32 IB Threads across 4 rigs 111 threads Crunching!!
hehe
[SIGPIC][/SIGPIC]
Have been going though valids and pendings at WCG comparing the CPU Time / Elapsed Time on the 2 PCs and since I made the priority change the numbers have gotten better and are better than the one with no change. The difference is small but just changing the priority using cc_config.xml on a dedicated cruncher seems fine. When I ssh in I can see no difference in responsiveness. I would not do it on a machine I used for other things. I see no reason not to do it on a dedicated cruncher.
Still have no idea why my top results are different though. But even dropping from 19 to 10 in the nice column does shave off a fraction to a full second on a 2 hour work unit. A very few 2 seconds. I added the setting to the cc_config.xml on the other PC as well.
My power bill came so I'm back to one desktop, but it'd be handy to bring this back up for next SC Week so everyone can add it to their headless farms.
[SIGPIC][/SIGPIC]
Well I dunno. Got up today and the machine I added the cc_config.xml setting to last night had set the nice value back to 19 on new work units started while I slumbered. So I got firm and did a renice. Worked for about 15-20 and it is renicing them back to 19.
The machine I had first reniced also had some CEP2 work units running at 19. So I reniced it again with a firm had and steely gaze. So far it is holding. I suspect some devilry.
OC maybe I am not the only one with memory issues. Are you sure you did not do something else? Doest thou dabble in the dark arts? Did you attend Hogwarts?
Last edited by PoppaGeek; 11-15-2013 at 07:18 AM.
Oh how I wish she cared what distro I used.
Poppa,
You better go for the strawberry parfait, the young lady might just reset your nice factor to 60 or 70....!!!
Nope, I did nothing else. I asked my cat and she agrees
None of mine run as a service, they are all found in /home/BOINC, other than that I am using Mint 14 cinnamon that has been dist-upgraded about 4-6 weeks ago and Aracnid is running 3.12 rc5 kernel.
Can't think that any of these details would be important but.....
Last edited by OldChap; 11-15-2013 at 09:47 AM.
My Biggest Fear Is When I die, My Wife Sells All My Stuff For What I Told Her I Paid For It.79 SB threads and 32 IB Threads across 4 rigs 111 threads Crunching!!
You may have answered my question. I am using the repositories startup scripts so they might be setting the nice there as it can be set via a command line switch.Nope, I did nothing else. I asked my cat and she agrees
None of mine run as a service, they are all found in /home/BOINC, other than that I am using Mint 14 cinnamon that has been dist-upgraded about 4-6 weeks ago and Aracnid is running 3.12 rc5 kernel.
Can't think that any of these details would be important but.....
EDIT:
Bingo!
Using Linux renice command not a boinc command line switch.# BOINC client needs higher priority than scientific apps
renice 10 -p $pid > /dev/null || renice -n 10 -p $pid
# Apps receive minimal priority
if [ -n "$children" ]; then
(chrt_many --idle 0 19 $children >/dev/null \
&& log_progress_msg "idleprio") || \
(chrt_many --batch 0 19 $children >/dev/null \
&& log_progress_msg "batch") || \
(chrt_many --other 0 19 $children >/dev/null \
&& log_progress_msg "normal") || \
log_progress_msg "chrt failed"
Last edited by PoppaGeek; 11-15-2013 at 10:05 AM.
Just like OC, my boinc is in /home/BOINC. But on Mint 13.
Yep that explains it. Since it is just a shell script it is easy to change so it is good to have figured it out.
So, to get things going the way we have them you remove or comment out the renice references ??
My Biggest Fear Is When I die, My Wife Sells All My Stuff For What I Told Her I Paid For It.79 SB threads and 32 IB Threads across 4 rigs 111 threads Crunching!!
OK here is the relevant section from the /etc/init.d/boinc-client file that starts boinc-client on boot and when you type sudo service boinc-client start/stop/restart/statusCode:# BOINC client needs higher priority than scientific apps renice 0 -p $pid > /dev/null || renice -n 0 -p $pid # Apps receive minimal priority if [ -n "$children" ]; then (chrt_many --idle 0 10 $children >/dev/null \ && log_progress_msg "idleprio") || \ (chrt_many --batch 0 10 $children >/dev/null \ && log_progress_msg "batch") || \ (chrt_many --other 0 10 $children >/dev/null \ && log_progress_msg "normal") || \ log_progress_msg "chrt failed"
NOTICE it says the client needs higher priority than the science apps. SO I changed renice 10 -p $pid > /dev/null || renice -n 10 -p $pid to renice 0 -p $pid > /dev/null || renice -n 0 -p $pid and everywhere it says chrt_many 0 19 $children to 10.
CEP2 still comes up at 19 but all others are 0, this with <no_change_priority> set to 1. Set it to 0 and everything comes up at 19.
EDIT:
The more I think about it I think Boinc client only needs a higher priority on a busy machine. On a full time cruncher with nothing else running, user stuff, I do not see why it would matter.
Last edited by PoppaGeek; 11-15-2013 at 11:14 AM.
and yet I have seen the benefit. I only took one snapshot for a different purpose that shows the worst case in action and, right at the bottom, what was then more normal for the remaining rigs. nothing else was running on any of them.
What we are looking at then is two different schedulers at work..... and one is better than the other.
before:
and after:
Notice the cpu percentage and the cpu time/real time on each
on screen for the first pic may be poor so direct links:
http://www.lakecityquietpills.com/ph...3194693020.png
http://www.lakecityquietpills.com/ph...8572620597.png
Last edited by OldChap; 11-15-2013 at 11:43 AM.
My Biggest Fear Is When I die, My Wife Sells All My Stuff For What I Told Her I Paid For It.79 SB threads and 32 IB Threads across 4 rigs 111 threads Crunching!!
I come from a production environment. If a system becomes unresponsive people were not getting their work done. If a system became unavailable the company lost thousands of dollars per MINUTE. From my perspective trying to save seconds by extreme overclocking or tweaking priorities does little good if it brings the machine down while I am not there to notice and get it going again and then loosing maybe hours.
While a few seconds here and there do add up over time they are easily lost with a machine down for even a few hours. Will playing with the "nice" setting cause unstabilty or lost time? Does not look like it so far but there are still things going on that no one here so far has an explanation for. So I don't know.
This is from my perspective and that is how I judge. Does not mean it is right for you. You should do whatever you feel is best and hopefully it will prove to have been the best for you. I will always choose the middle path, mild overclock with mild temps and a bump in priority. Hopefully it will prove to have been the best choice for me.
I still want to know what causes CEP2 to run at a lower priority and whey changing settings does not always take. Linux seldom needs a reboot as Windows does. Usually just a kernel change so that is not it.
It may be that there are two competing systems at work but I see CFS as the way it should be done if all processes are equal and nice interrupts this by changing the process priority for certain things.
My take on this is that for a dedicated rig you should run cfs but if the rig is multi use you should attempt to run cfs and be prepared to revert to setting nice if you encounter unresponsiveness.
http://en.wikipedia.org/wiki/Completely_Fair_Scheduler
My Biggest Fear Is When I die, My Wife Sells All My Stuff For What I Told Her I Paid For It.79 SB threads and 32 IB Threads across 4 rigs 111 threads Crunching!!
Are you running CFS? I thought you were using <no_change_priority>?
I feel like I have missed something again....
I'll read about it later. Eyes are bleeding from reading config files and error logs.
Posting Permissions
Reply With Quote
Bookmarks