- WLUG - wlug.mailman3.com

Reminder: Meeting Wed 14-dec-05 7:00 PM

by Andy Stewart

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 HI Everybody, Our next meeting is coming up really soon - this Wednesday, Dec 14, at 7:00 PM in the Access Grid room at WPI (lower level of Morgan Hall). I don't have any defined topic for the meeting, so I'm open to any ideas you might have. I have some flyers from Addison-Wesley & Prentice Hall which are for a 40% (yes, forty percent) discount on books purchased online prior to January 10, 2006. They will be available at this meeting. Also, I'll bring the following books for the zero cost raffle: Linux, Debugging and Performance Tuning Author: Steve Best Prentice Hall Apache Derby - Off to the Races Includes Details of IBM Cloudscape Authors: Paul C. Zikopoulos, Dan Scott, George Baklarz IBM Press That's about it - I look forward to seeing you at the meeting! Andy - -- Andy Stewart, Founder Worcester Linux Users' Group Worcester, MA, USA http://www.wlug.org -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.2.5 (GNU/Linux) Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org iD8DBQFDnkIzHl0iXDssISsRAiR8AJ9D/L0X/Iak0TLPTt0dnA6Hgg288ACeMYdn zWNMvvBXjO7C9Cq6zSuPNH0= =enyh -----END PGP SIGNATURE-----

Dec. 13, 2005

Re: [Wlug] getting that cpu to sleep ...

by Keith Wright

> From: brad <maitre(a)ccs.neu.edu> > > > The immediate task at hand is that i want to get readings from > an instrument (a light sensor) at a uniform interval, say every > 1000 micro seconds. I have been able to achieve this task by > setting up the loop i described above. However the loop above > takes 99% of the CPU. I'd like to be able to achieve this > without pegging the CPU at 99%. Does this make sense? if not i > can try explaining it again, perhaps with some code. > > The kernel i'm running is 2.6.12. (well uname -r reports > '2.6.12-9-686-smp.' its ubuntu's stock smp kernel.) You are doing something that Linux was not designed to do. But don't despair, you are not the first one. There is RT-Linux, (Real-Time Linux), which consists of a real-time scheduler (meaning that it is designed to schedule tasks with non-negotiable time constraints), which treats the entire Linux kernel as a background task. Assuming your raw processor speed is sufficient (without which you lose no matter what you do), the RT-Linux scheduler will run your special task at guaranteed intervals, while any network, disk I/O, web browsing, etc. that you need to do can use the Linux system in the cracks between. Also the embedded Linux project and uCLinux (microComputer) Linux might be interesting. Give any of these terms to Google and get back more than you can read, but here is a link to get started: http://www.linuxdevices.com/articles/AT3694406595.html -- Keith

Dec. 13, 2005

Re: [Wlug] getting that cpu to sleep ...

by brad

On Mon, Dec 12, 2005 at 06:17:35PM -0500, Jeff Moyer wrote: > ==> Regarding [Wlug] getting that cpu to sleep ...; brad <maitre(a)ccs.neu.edu> adds: > > maitre> greetings, > > maitre> I have a question for those linux programmers out there. i'm > maitre> writing a program on linux (kubuntu to be specific) that requires > maitre> high-precision timing, somewhere on the order of 100 to 1000 micro > maitre> seconds. I get very good timing out of a loop written such that a > maitre> while() loop checks the system clock after every iteration and > maitre> comparing that to the desired wait time, like so (pseudo_code): > maitre> while(time_passed < wait_time) { time_passed = get_time() } ... > > maitre> The problem with this approach is that the process uses 99% of the > maitre> CPU since it just spins the CPU until enough time has expired. My > > The other problem with this is that you can be scheduled if your timeslice > runs out. > ummm, i didn't think about that one. > maitre> question: is there a way on linux (in C) to have the CPU sleep or > maitre> send it a no-op such that it's not just eating up cycles until time > maitre> expires? I have looked at nanosleep and usleep, however they appear > maitre> to only have precision to 10ms which too latent for my needs. > > As Mike pointed out, in more recent kernels you can tune the value of HZ to > get better granularity from select/poll. > Actually Mike's response reminded me of an article i bumped into about nanosleep on dragonflyBSD http://www.dragonflybsd.org/docs/nanosleep/ > What, precisely, are you trying to do and on what kernel? Perhaps with > more information we could provide better guidance. > The immediate task at hand is that i want to get readings from an instrument (a light sensor) at a uniform interval, say every 1000 micro seconds. I have been able to achieve this task by setting up the loop i described above. However the loop above takes 99% of the CPU. I'd like to be able to achieve this without pegging the CPU at 99%. Does this make sense? if not i can try explaining it again, perhaps with some code. The kernel i'm running is 2.6.12. (well uname -r reports '2.6.12-9-686-smp.' its ubuntu's stock smp kernel.) -brad

Dec. 13, 2005

Re: [Wlug] getting that cpu to sleep ...

by Bradley Noyes

On Mon, Dec 12, 2005 at 06:17:35PM -0500, Jeff Moyer wrote: > ==> Regarding [Wlug] getting that cpu to sleep ...; brad <maitre(a)ccs.neu.edu> adds: > > maitre> greetings, > > maitre> I have a question for those linux programmers out there. i'm > maitre> writing a program on linux (kubuntu to be specific) that requires > maitre> high-precision timing, somewhere on the order of 100 to 1000 micro > maitre> seconds. I get very good timing out of a loop written such that a > maitre> while() loop checks the system clock after every iteration and > maitre> comparing that to the desired wait time, like so (pseudo_code): > maitre> while(time_passed < wait_time) { time_passed = get_time() } ... > > maitre> The problem with this approach is that the process uses 99% of the > maitre> CPU since it just spins the CPU until enough time has expired. My > > The other problem with this is that you can be scheduled if your timeslice > runs out. > umm, i didn't think about that one. > maitre> question: is there a way on linux (in C) to have the CPU sleep or > maitre> send it a no-op such that it's not just eating up cycles until time > maitre> expires? I have looked at nanosleep and usleep, however they appear > maitre> to only have precision to 10ms which too latent for my needs. > > As Mike pointed out, in more recent kernels you can tune the value of HZ to > get better granularity from select/poll. > Actually Mike's response reminded me of an article i bumped into about nanosleep on dragonflyBSD http://www.dragonflybsd.org/docs/nanosleep/ > What, precisely, are you trying to do and on what kernel? Perhaps with > more information we could provide better guidance. > The immediate task at hand is that i want to get readings from an instrument (a light sensor) at a uniform interval, say every 1000 micro seconds. I have been able to achieve this task by setting up the loop i described above. However the loop above takes 99% of the CPU. I'd like to be able to achieve this without pegging the CPU at 99%. Does this make sense? if not i can try explaining it again, perhaps with some code. The kernel i'm running is 2.6.12. (well uname -r reports '2.6.12-9-686-smp.' its ubuntu's stock smp kernel.) -brad

Dec. 13, 2005

Re: [Wlug] getting that cpu to sleep ...

by Jeff Moyer

==> Regarding [Wlug] getting that cpu to sleep ...; brad <maitre(a)ccs.neu.edu> adds: maitre> greetings, maitre> I have a question for those linux programmers out there. i'm maitre> writing a program on linux (kubuntu to be specific) that requires maitre> high-precision timing, somewhere on the order of 100 to 1000 micro maitre> seconds. I get very good timing out of a loop written such that a maitre> while() loop checks the system clock after every iteration and maitre> comparing that to the desired wait time, like so (pseudo_code): maitre> while(time_passed < wait_time) { time_passed = get_time() } ... maitre> The problem with this approach is that the process uses 99% of the maitre> CPU since it just spins the CPU until enough time has expired. My The other problem with this is that you can be scheduled if your timeslice runs out. maitre> question: is there a way on linux (in C) to have the CPU sleep or maitre> send it a no-op such that it's not just eating up cycles until time maitre> expires? I have looked at nanosleep and usleep, however they appear maitre> to only have precision to 10ms which too latent for my needs. As Mike pointed out, in more recent kernels you can tune the value of HZ to get better granularity from select/poll. What, precisely, are you trying to do and on what kernel? Perhaps with more information we could provide better guidance. -Jeff

Dec. 12, 2005

Re: [Wlug] getting that cpu to sleep ...

by Mike Frysinger

On Mon, Dec 12, 2005 at 05:48:24PM -0500, brad wrote: > My question: is here a way on linux (in C) to have the CPU sleep or > send it a no-op such that it's not just eating up cycles until time > expires? I have looked at nanosleep and usleep, however they appear > to only have precision to 10ms which too latent for my needs. you cant unless you tweak the kernel HZ settings read this page: http://kerneltrap.org/node/464 -mike

Dec. 12, 2005

getting that cpu to sleep ...

by brad

greetings, I have a question for those linux programmers out there. i'm writing a program on linux (kubuntu to be specific) that requires high-precision timing, somewhere on the order of 100 to 1000 micro seconds. I get very good timing out of a loop written such that a while() loop checks the system clock after every iteration and comparing that to the desired wait time, like so (pseudo_code): while(time_passed < wait_time) { time_passed = get_time() } ... The problem with this approach is that the process uses 99% of the CPU since it just spins the CPU until enough time has expired. My question: is there a way on linux (in C) to have the CPU sleep or send it a no-op such that it's not just eating up cycles until time expires? I have looked at nanosleep and usleep, however they appear to only have precision to 10ms which too latent for my needs. Thanks, --Brad

Dec. 12, 2005

Success, almost, with GNOME

by Ken Jones

Per the suggestion of Chuck Anderson I logged in under a new account. GNOME desktop worked just as it should. So I removed all of the /home/kjones/.gconf directory and logged in again as myself. Still it did not work correctly. In particular the right click on empty space did not work. I did a "diff -r" on the two (now default versions I suppose) .gconf directories. They seemed the same save for unique identifiers. Where do I look to find the differences? One other question. The Applications launcher on GNOME does not contain the LOG OUT feature. I know the command "shutdown", but I can not find a command to Log Out. Is there one? What is it? Ken Jones

Dec. 12, 2005

Re: [Wlug] Problem with gnome, Fedora core 3

by Chuck Anderson

On Sat, Dec 10, 2005 at 03:58:05PM -0500, Ken Jones wrote: > Any ideas what controls this behavior and ow I can fix it? On a > different machine I have a Core4 system running GNOME perfectly. Try creating a new account and logging in to that one to see if the problem exists there. If not, then it is definately something in your user account, such as your gconf settings.

Dec. 10, 2005

Problem with gnome, Fedora core 3

by Ken Jones

Gentlemen and ladies, My Fedora Core 3 - which has been yum'd up to date - behaves strangely when using GNOME as the windows manager. The right click on blank desktop space does not work. The right click on a panel to make a new panel does not work. The result is that I can not set up any launching icons on my desktop. There is no unused space on my one (bottom) panel. The panel-disappear icon is missing. Any ideas what controls this behavior and ow I can fix it? On a different machine I have a Core4 system running GNOME perfectly. Ken Jones

Dec. 10, 2005