The Linux Help and Tutorial Thread

[D_A]

Found this on the Ubuntu forums. Looks like it should help.

http://ubuntuforums.org/showthread.php?t=182764

1. Set up your /boot/grub/menu.lst to change the size of ramdisks
2. Make a filesystem on the ramdisk
3. Mount the ramdisk somewhere

STEP 1.

Open a terminal and type:
Code:

sudo gedit /boot/grub/menu.lst

Enter your password to be able to edit the file

Find a section that looks something like this:
Code:

title Ubuntu Linux
root (hd0,4)
kernel /boot/vmlinuz-2.6.12-10-686 root=/dev/hda5 ro quiet splash
initrd /boot/initrd.img-2.6.12-10-686
savedefault
boot

Add ramdisk_size=500000 to the end of the kernel line, replacing 500000 with whatever size you want your ramdisk to be in KB (this example makes a 500mb drive)

After modification, the line should look like this:

Code:

kernel /boot/vmlinuz-2.6.12-10-686 root=/dev/hda5 ro quiet splash ramdisk_size=500000

You need to reboot in order for the change to the ramdisk size to take place, otherwise you'll end up with a smaller (around 50mb for me initially) ramdisk.
The ramdisk size seems to be passed to the kernel at boot time, and all of /dev/ram* end up being that size.

STEP 2.

You need to make a filesystem on /dev/ram0 in order to be able to put files on it.

In a terminal, type:
Code:

sudo mke2fs /dev/ram0

STEP 3.

Now, you need to mount the filesystem somewhere.
Again, in a terminal, type:
Code:


cd /media/
sudo mkdir ramdisk
sudo mount /dev/ram0 /media/ramdisk/
sudo chmod -R 777 ramdisk/




Thats it. now you have a ramdisk formatted with ext2!

[D_A]

I'm getting real curious if this will have a substantial impact on BOINC performance, on how hard CEP2 works the drives and if it would be beneficial for SSDs as well.

[Gamekiller]

I'm getting real curious if this will have a substantial impact on BOINC performance, on how hard CEP2 works the drives and if it would be beneficial for SSDs as well.

I believe it does help. There were a few threads on the WCG forums about it and on these forums as well.

Found this on the Ubuntu forums. Looks like it should help.

http://ubuntuforums.org/showthread.php?t=182764

I tried that and couldn't find /boot/grub/menu.lst (didn't exist) so I figure it was old and stopped reading.

Edit: Tried:
sudo mke2fs /tmp/MC
in reply i got: "/tmp/MC is not a block special device. Proceed anyway? (y,n)"

[D_A]

This might be helpful: http://superuser.com/questions/17586...k-ubuntu-10-04

If I'm getting this, all you need to do is specific one line in your fstab and you're good to go. I'll have a little play with this and see what happens.

[D_A]

Scratch that, didn't work as hoped. Will keep playing.

[bearcatrp]

Do not use the System Monitor. It will show show your CPUs at all sorts of percentage of use. The System Monitor runs at a higher priority than Boinc WUs. Open a Terminal and type top. It shows more info anyway just not as pretty.

Ran top. About the same as system monitor. Have usually 3 or 4 dipping below 40%. Never all at 100% or even close. Since I can't monitor my temps, am wondering if something is getting hot enough to trigger ubuntu or the bios to throttle.

[PoppaGeek]

Ran top. About the same as system monitor. Have usually 3 or 4 dipping below 40%. Never all at 100% or even close. Since I can't monitor my temps, am wondering if something is getting hot enough to trigger ubuntu or the bios to throttle.

In TOP in the %CPU column is anything else showing more than a few %? What do Load Averages show in top right corner?

[D_A]

Ubuntu won't throttle your CPU unless you specifically tell it to. Your BIOS defaults assume you WANT it to.

[josh1980]

Ok, in case anyone hasn't been reading the GPUGRID forum, I have issues running WCG and GPUGRID at the same time. It seems that the GPUGRID task get's choked out of resources. My dedicated linux box currently has a task duration correction factor of 8.7. It was 11 3 days ago until I suspended WCG and started goofing around. I've found a very cool command, renice. If I change the priority default for WCG tasks (19, very low priority) to normal priority (0) then the GPUGRID begins to crunch at a much more expected rate. So here's my question...

How do I set up a command to run every 20 minutes or so that can automatically change the priority of all processes with a name of acemd2* to have a priority of normal?

I found the renice command and I'm sure renice does exactly what I want, but I have no idea how to set up a scheduled task to execute a renice command every 20 minutes. Is there a script someone has that can do this? I've spent 2 hours trying to find scripts that do similar things so I could cannibal them and figure it out on my own, but I've run out of search terms to use. As it stands, my linux box is outputting just 13k/day average with a GTX 275.

Thanks!

[D_A]

I know it's possible but I'm not that good at script writing. There's an app on source forge called "niceall" that lets you renice multiple related threads with one command but I haven't played with it. You'd have to compile it yourself, too.

[PoppaGeek]

Write a script to execute the command in crontab as I suggested to GK here

[bearcatrp]

In TOP in the %CPU column is anything else showing more than a few %? What do Load Averages show in top right corner?

here is a screen shot

[PoppaGeek]

Load averages are no help as system has only been up 17 minutes. All the 90% and up are ok but there are alot of below 90%. I am curious what it would show after being up for an hour or more. Also close the Boinc Manager. Processes bounce around to different CPUs and it is possible that Xorg and Boinc Manager are most of the cause of the lower CPU utilization but it does look worse than I would expect. Compiling your own kernel might improve it. To my knowledge Linux does not throttle CPU due to temps. It is usually hard to even get a good temp reading in Linux.


EDIT:
Since you seem to be running all C4CW and the work units are pretty consistent have a look at the completed WUs for that machine and see if there are big differences in run time. If you have another machine running them also compare times or poet times here along with the CPU speed and we can get an idea if what we are seeing in Top is a problem or not.

[bearcatrp]

My other 2 crunchers do not have this problem. Harpertown and woodcrest xeons. They are pegged at 100% on all cores. Would hyperthreading be the problem? I did shut that off to see what would happen and it still does this. Me and Dave went through everything in the bios too. Doesn't do this in win7. It's got to be ubuntu doing this. Will stick with ubuntu as it get around 112 wu's in 24 hours versus 80 under win7.

[PoppaGeek]

Ubuntu may be doing it but I do not see how it could be because of temperature.

How does the complete time on WUs compare?

At command line type exactly as shown:

mpstat -P ALL 2 5

If the software is not installed it will tell you how to. What is the idle time of the CPUs? does it match what you see in top? In other words top shows a CPU at 60% load, does mpstat show it 40% idle? If it shows the CPUs at 0% idle then something is missreporting the CPU usage or something is stealing the CPU time.



If all this is getting on your nerves then do not run top and leave it alone. It is still doing better on C4CW than Win7.

[bearcatrp]

here you go...

[D_A]

Ok, to clarify: Dies this machine have 16 physical cores or 8 physical and the other 8 from hyperthreading?

[PoppaGeek]

Have you checked the preferences in Boinc Manager for Processor Usage? Set to use 100% of processors 100% of the time?

I run all AMD and know nothing about Intel so I am no help there and I am running out of ideas......

[bearcatrp]

2 e5640's. 8 cores, 16 threads. Everything is set to 100 percent. Am going through the bios again and changing one thing at a time. My other crunchers use asus boards. This one uses super micro. This is really bugging me.

[D_A]

If that's 8 physical cores with hyperthreading then those figures are probably accurate. The virtual cores created by hyperthreading don't give you a 100% performance boost since they're not real cores. What you're seeing there isn't your system losing 40% of crunching time, what you're seeing is you're system GAINING 60% (and more) crunching performance due to hyperthreading.

[bearcatrp]

Weird it does this in ubuntu but not in win7 but since ubuntu spanks win7, will keep running ubuntu. That's folks for the help.

[PoppaGeek]

Maybe it is just the way CPU usage is calculated in Windows and Linux.

[Gamekiller]

Sadly I think I'll be moving my server over to windows. Having to search for a fix for every simple action for 8 hours a day was fun for a little while but I'm starting to grow tired of it.

Thanks for all of the help you guys have given me thus far.

[PoppaGeek]

Yes it gets old. There are a lot of things you can do with Linux but sometimes it seems more painful that it is worth to get there.

[Bobsama]

So I have a Pentium E2140 @ 2.4GHz with 1GB RAM running Ubuntu 9.04 for AMD64. It's exclusively crunching HCC. http://boincstats.com/stats/host_gra...wcg&id=1345004

RAC(WCG): 1092.995...

Mind you, the WCG PPD Estimator v0.39b says a Core 2 Duo (Conroe) at 2.4GHz should get 754.29 BOINC/day. Not bad to be beating that number by 45%.

I'm thinking of moving the main over to the same OS (via virtual machine) since it seems to be crunching so damn well. On the other hand, my Pentium D's output has dropped on the same regiment (HCC), even with its CPU cooling upgrade. It went from roughly 418 BOINC/day to 329 BOINC/day. That machine especially I've been thinking of changing over to Linux (it's just a file server--but I can't figure out how to move the computer over to "WORKGROUP").