The Linux Help and Tutorial Thread

Several years ago if you opened System Manager while running Boinc work units the CPU graphs went wild and CPU% would jump wildly. That does not seem to be the case now. I have checked it on the 2 Ubuntu boxes I have a GUI on and it acts fine. Some things improve with age.

Feel free to use System Manager with reckless abandon.;)

New kernel! DIY

Code:

---

wget https://www.kernel.org/pub/linux/kernel/v3.x/linux-3.12.tar.xz

---

Lets download this brand new stable mainline kernel - https://www.kernel.org/ (for all options)

Code:

---

sudo tar -xvf linux-3.12.tar.xz -C /usr/src/

---

..and extract it to /usr/src

Code:

---

cd /usr/src/linux-3.12/

---

..and go there

Code:

---

sudo make localmodconfig

---

lets remove EVERYTHING that is not in Your computer, or rather, build the kernel for this specific HW configuration - guess this is most reasonable for cruncher

Code:

---

sudo make -j13 deb-pkg

---

..since my opty has 12c I can put them all working - its available cores/threads +1

Code:

---

cd /usr/src/

---

up one level

Code:

---

sudo dpkg -i linux-image-3.12.0[TAB]

---

lets deploy that freshly baked niceness, [TAB] press the button here

Code:

---

sudo dpkg -i linux-headers-3.12.0[TAB]

---

that too

Code:

---

sudo shutdown -r now

---

time for restart!

Now, on boot options screen - select the new kernel from "advanced options".

Profit - if its not fast or has some issues then you can always revert to previous, again on "advanced options" on boot.

[the above is in no way 100% correct way, as im also a fresh convert to linux - but i think/hope any errors will be pointed out]

The above has been tested by me on my cruncher, which is still alive and has gained (with personalized kernel/frequent updates) ~15% (6,5K to 7,5K avg ppd)

If all is good with the new kernel - then lets remove the old one(s)

Code:

---

dpkg --list | grep linux-image

---

Gives the list of currently in the system

Code:

---

sudo apt-get purge linux-image-[insert old kernel number here]

---

Remove old ones not needed

Code:

---

sudo update-grub2

---

Update bootmenu

Profit!

question: why would I see BoincTasks showing 94% (typical) for each wu when top shows 15*100% and the other 98% in use by wu's

I have watched without boinc running and seen maybe 8-10% of a thread used by other processes.

also, in an ideal world, I would expect to see (again with top) "user" at a v. high percentage and "nice" very low %. I actually see the opposite. Can we alter the cpu priorities within boinc to "normal" and maybe let "nice" do all that work because just now there seems to be a clash between the two systems.

I get ~99,9% for MCM

http://i.imgur.com/ZbPNxlV.jpg

http://i.imgur.com/SrZO3gT.jpg

I get (got) 94.6% on one rig which I just re-booted to have a look at what was going on... meanwhile I have that work suspended and a bunch of high priority work started which is running at 99.46-99.89% :shrug:

Just noticed on another rig a wu at 47% cpu at 99.81% and runtime so far of 13hrs 50mins :eek:

Lots of variance for these WUs.

Seem to run ok though, for me. CPU temps are the same.

Quote:

---

Originally Posted by OldChap

Can we alter the cpu priorities within boinc to "normal" and maybe let "nice" do all that work because just now there seems to be a clash between the two systems.

---

There is some thing called "nice level". Looks like you can change that in the cc_config file.

Edit: forgot to add the link.

http://boinc.berkeley.edu/wiki/Client_configuration

Look for "no_priority_change"

Quote:

---

Originally Posted by Rob_B

Look for "no_priority_change"

---

lol... and i've had cron script to do this :rofl:

should RTFM before i do things the hard way

Quote:

---

Originally Posted by OldChap

question: why would I see BoincTasks showing 94% (typical) for each wu when top shows 15*100% and the other 98% in use by wu's

I have watched without boinc running and seen maybe 8-10% of a thread used by other processes.

also, in an ideal world, I would expect to see (again with top) "user" at a v. high percentage and "nice" very low %. I actually see the opposite. Can we alter the cpu priorities within boinc to "normal" and maybe let "nice" do all that work because just now there seems to be a clash between the two systems.

---

"nice" is the priority that BOINC tasks run at, which is why you see so much CPU usage in that category. Threads with a "nice" priority stop to allow other threads to use the CPU, which is how BOINC gets out of the way to allow you to use your system. "User" priority threads are those that you start deliberately, such as your browser, email client, image viewer etc. If you jack the BOINC process priority up to "user" level you will find the system much less responsive to your own tasks and usage and harder to administer.
Strictly speaking, the whole thing is called "nice" and BOINC worker tasks have a "nice" of 20, meaning they are very nice and get out of the way of everything, while "user" tasks have a lower number, meaning a higher priority, and they are more selfish. Tasks listed as "nice" in TOP etc are just one step higher priority than "idle".

I tried it on one rig and looks like it changes some thing. Only the CEP2 is still on the lower priority. :confused:

http://www.robbakker.com/images/top1.png

Well first of all I had not seen that before so thanks Rob.

D_A: I found the following comment somewhat edifying: Nice is helpful when you are competing against other users on a single machine, but for a single user with multiple jobs, let the scheduler do the scheduling.

I have changed mine now on the dedicated crunchers... just to see what happens. I am not expecting a lot but who knows

EDIT: I can't get screenshot working but I now have 99.8or9% user and nil nice on the one rig I checked so far

http://img.techpowerup.org/131111/1111000000.jpg

They are all Linux. I have noticed many times with BoincTasks that when the numbers are low and I watch it for awhile they come up slowly. I guess it does it's own averaging or some scoccery I do not know. Top is what you should go by. X, the GUI does take some CPU so if you are using the machine moving windows around, opening and closing them it will take some from Boinc tasks.

Installed Mint 64 Cinnamon. I can see why people like it. I was a little apprehensive when I was not given the opportunity to use parted, I may have missed it. Pretty neat trick doing a side by side install. I was pretty aggravated at first as the menu was choppy and jerky but it was the theme I was using, others were fine. Copied /var/lib/boinc-client over from the Ubuntu install after installing from repositories and copying boinc* from the 7.0.65 archive. Ran into ownership problems. Solved them with

Quote:

---

sudo chown -R boinc:boinc *

---

while in /var/lib/boinc-client/

Picked up where it left off.


BTW - when copying the directories do not worry about errors about the sym links. They were all made when you install from repositories: sudo apt-get install boinc-client. :up:

Yes, you're right. You missed it. You can partition manually if you want, it's the bottom option of the same page where you chose to install side-by-side.

More about "nice"

My linux install on my "Spiders" rig was new and that has seemed to perform very well anyway but the longest wu that I have yet run is just that bit better with nice off and the OS scheduler doing the scheduling. (don't ask me which scheduler is in use here)

http://www.lakecityquietpills.com/ph...9269332101.png

The loss then is 0.07 of an hour in 29 hours = ~4 mins :)

My Arachnid rig is using a linux install that was previously in use on another rig. It was doing OK running in the 97% region. not so great but again changing Nice has now put this into the 99% range.

No pic this time but has lost 58 seconds in 5 hours 13 mins on the current longest wu.

Then we get to the rig with the single processor. This is an HDD that has gone from rig to rig, has been updated, upgraded and generally needs a fresh start it seems. This one continues to run in the 94% range with 16 cores.

switching to 15 cores running pushes the performance back into the 99% range but even though those 15 cores show at 100% and there is no other cpu use seen in top beyond the odd 1-2% used by sys and cinnamon there is currently only 6% showing idle.

This one is going to need work (even if only a fresh install etc)

@OC How did you change the nice value? Can you show a screenshot of top running showing all CPUs? With top running hit 1. PLease. I am experimenting as well using the cc_config.xml way and am not seeing any difference in the %ni or %id columns.

Thanks
---------------------------------------------------------

The top man page does not define all the columns when you hit 1 to see all CPUs.

us: user cpu time (or) % CPU time spent in user space
sy: system cpu time (or) % CPU time spent in kernel space
ni: user nice cpu time (or) % CPU time spent on low priority processes
id: idle cpu time (or) % CPU time spent idle
wa: io wait cpu time (or) % CPU time spent in wait (on disk)
hi: hardware irq (or) % CPU time spent servicing/handling hardware interrupts
si: software irq (or) % CPU time spent servicing/handling software interrupts
st: steal time - - % CPU time in involuntary wait by virtual cpu while hypervisor is servicing another processor (or) % CPU time stolen from a virtual machine

http://img.techpowerup.org/131112/_0000000.jpg

The top one is with <no_priority_change> undefined. The lower one is with <no_priority_change>1</no_priority_change>

The top one is running 2 CEP2 work units, only time I ever really see anything in the %sy column over .ox, or more than less than 1%.

Not seeing criticism but we are possibly talking at crossed purposes.

When I am talking about seeing 94% I am referring to the Boinctasks screen which, oddly enough, I am tending to go with because it seems to accurately fit with the amount of lost time (the difference between real time and cpu time for a wu.)

All the while there is 94% showing in Boinctasks there is 100% showing on all bar one thread (98%) in top.

When I was referring to the idle time, that was seen in the info across the top of top: user, system, nice, idle, wait etc.

All together this is saying to me that if it is not the OS being inefficient, seemingly the case judging by the 100% for each thread showing but there is only 6% of idle time when running 15 threads then the losses must lay elsewhere and perhaps a new install would rule out less than optimum "drivers" etc.

It could be lesser hardware causing this but....

I am just following my nose on this and my logic may be faulty too bearing in mind my lack of detailed knowledge on how this stuff actually works..... but it should only cost an hour to find out.

EDIT: I have an issue trying to get a screenshot... what I get just now is a screen that is older than the current screen. any tips on that and I will gladly post.

I changed my value using boinktasks> extra> edit config file> <no_priority_change>1</no_priority_change>

I tried just getting it to read the config file again but had to actually re-start boinc for it to work

You could do this with gedit and just add the above line to cc_config. (listen to me talking to the bearded one ....) :rofl:

I was wondering if any of you Linux guys who have switched over to the newest BOINC client for 64 bit Linux (7.2.28?) have seen a difference in crunching times as compared to 7.0.65 or .64?


Also, as a Linux newb, I am still learning things about the OS. To that end, did you or do you have to use a command line way of updating to the new version or did you have to uninstall the current BOINC client and then install the new one?

sorry OC I changed my post above. The more I got to thinking about it all I thought I did not really understand things as well as I thought. Still not sure. Why are all processes showing %id when they show no reason for it. no IO wait states, no system or %sy, except for CEP2. I used to have to deal with this stuff on very busy database servers with hundreds of users. Not sure it this is all different, Linux is that different than Solaris in how it reports these things, I am old and with foggy brain and beginnings of dementia. Maybe some of all three.

Will be interesting to see what we learn. :up:

Quote:

---

Originally Posted by yojimbo197

I was wondering if any of you Linux guys who have switched over to the newest BOINC client for 64 bit Linux (7.2.28?) have seen a difference in crunching times as compared to 7.0.65 or .64?


Also, as a Linux newb, I am still learning things about the OS. To that end, did you or do you have to use a command line way of updating to the new version or did you have to uninstall the current BOINC client and then install the new one?

---

I have seen no difference but the newer one allows you to use app_config.xml

see this thread on how to update and or install newest Boinc.

Jim, Command line is the easiest way I know Poppa posted at start of his update linux thread or there is the D_A post on the using Linux thread

http://www.xtremesystems.org/forums/...ely-site-Linux

http://www.xtremesystems.org/forums/...=1#post5024891

whichever you like.

hard to tell if any improvement due to change to MCM1

Poppa, edit to my previous too

Quote:

---

Originally Posted by PoppaGeek

sorry OC I changed my post above. The more I got to thinking about it all I thought I did not really understand things as well as I thought. Still not sure. Why are all processes showing %id when they show no reason for it. no IO wait states, no system or %sy, except for CEP2. I used to have to deal with this stuff on very busy database servers with hundreds of users. Not sure it this is all different, Linux is that different than Solaris in how it reports these things, I am old and with foggy brain and beginnings of dementia. Maybe some of all three.

Will be interesting to see what we learn. :up:

---

I'd have to look it up but all the reporting stuff in Solaris is available in the Linux version of top, you just need to use the right flags ... which I don't remember. I'm sure you can configure top's default to show whatever you want, but again I don't remember the syntax etc. man top is your friend.

When you run top as above do you see about the same thing?

I'll steal rob's ss for this:

http://www.lakecityquietpills.com/ph...4658855105.png

on mine 1. has gone to 90+%

2 and all in column 3= 0%

4. all 100%

If you have added the line to cc_config then quit and re-start Boinc and yours should change too

installed shutter

http://lakecityquietpills.com/photo/...1216932189.png

I think that all commands can use -h for a help syntax screen