SMP folding client uses one core on new kernel build

[Daniel ~]

In one way or another I feel like I've spent my entire computing life....waiting for another revision...it's true! LOL

[ccperf721p]

Originally Posted by Daniel ~

In one way or another I feel like I've spent my entire computing life....waiting for another revision...it's true! LOL

Then you find out the next revision hasn't changed or added anything you actually wanted..

[ccperf721p]

Originally Posted by ThunderRd

That's OK, it's more of an exercise in learning for me, I may stick with the one that's working well for me and wait for the next major revision to try again.

I wish I could open some dialog with the RT developers but they are a rather secretive bunch...I reckon that I'd get lost in the first few words anyway

As for the RT patch, I have found it to be noticeably lighter and quicker on its feet with desktop apps than the standard release kernel. It also is slightly faster in frame times in folding; don't know why that should be though.

I had to take a learning break. It seemed I was spending way too much time learning how to fix something that was working before version 3.8.23_rc765-2009187.1.3.5.9 (Joke) I also noticed that there has been a serious case of dependancies being dragged in with apps that I use that did not have hardly any dependancies before. Strange part is they do not offer any additinal functionality over my version.. I think they just keep hacking on the code until it is so bloated and useless they have to move on to version 2.0 and start over..

I don't enjoy talking to devs most of the time either.. I run into the same kind of conversation with a couple of the engine builders in this town.. Ask them why they did a certain thing and they want to act as if it is classified information.. Right up to the point where I tell them it did not work and his customer is now my customer..

[Daniel ~]

Originally Posted by ccperf721p

Then you find out the next revision hasn't changed or added anything you actually wanted..

And broke something you did!

[ccperf721p]

Originally Posted by Daniel ~

And broke something you did!

I got a good laugh out of that one. Too true.

If it's good and it works, it must not be finished yet.

I was reading on /. that the linux kernel was over 10 million lines of code and getting bigger with every release. This is before any distribution adds their patches and code. Not sure at what point it becomes bloated but I think it's getting there..

[Daniel ~]

Truth is I'm still so happy to be free of MS that I'm willing to put up with the stuff "They "
( Linux dev's ) think is fun to play with or needs an upgrade. The price is still right for me!! ":O}

[ccperf721p]

I'm still bound to Windows by my software providers and must pay them an anuual license fee, so the price of linux doesn't really matter. I enjoy linux for what it stands for more than anything else, but in my opinion it is heading a little too much in the Window-esque direction, without the backing of any major software being available to it.

I figure I have given MS about $2000 over the course of my computing time. Which is a very small percentage of what I have given other software companies, so I am unable to feel that I have been ripped off by MS..

[Daniel ~]

No one ever said it would be easy learning to compute! ":O}

[ccperf721p]

Originally Posted by Daniel ~

No one ever said it would be easy learning to compute! ":O}

That's for sure.. I haven't given up on putting my own package together. I've only had a couple of hours a week to work on it but it's getting close to at least booting to a command line.. I was unable to find a suitable base in another distribution so I decided to start from scratch. Maybe by the weekend I will have a kernel installed and can begin to test the base system.

[ccperf721p]

Hey TR. Are you patching a vanilla kernel or someone elses sources. It hadn't occurred to me yet, but I had seen big time problems when trying to add a patchset to a CK-sources kernel around 2.6.22.

[ThunderRd]

Vanilla-straight from Kernel.org.

Last night I was farting around and using the same base version patched it with the Zen patch...that seemed to run both cores fine using the same .config.

So it is definitely a setting to enable in the RT kernel patch or it's a bug. I don't know where to begin to report a bug of this nature; the disto forum hasn't had any responses, and there is no contact I can find with the RT developers other than the mailing list. I'm not inclined to bring my sorry-azzed-newbness to the mailing list so I reckon I'll just have to wait for the major revision and see if it still happens.

[ccperf721p]

I guess the spirit of linux isn't what it used to be. Can't believe there hasn't been anyone at least try to help. While I don't remember making the decision to stay away from experimental features and patchsets, I do remember using quite a few of them a few years back, I'm wondering if not being able to get help was one of the reasons I left it for dead.

I do remember being kicked from plenty of forums for speaking my mind about their paticular community. I would be especially upset when I would post in a thread as the new guy and it was like my post wasn't even there as the thread continued or treated as a noob when the things I had already tried were for more advanced than the suggestions being made.

I've downloaded the sources and the RT patch to my backup rig. I will install it tomorrow night using as much of your config as I can. I can come close but it will be a core2 instead of amd64. Only amd64 rig I have left is the dual opty rig I run my business on so can't go playing with that..

[ThunderRd]

Cliff-
I have trimmed down the .config so that the initrd.img is under 5MB. I figured, if I can't get the newer kernel to run I'd might as well lean out the old one while I was inclined and while I had my head into it.

I'll attach the .config I'm using now, but don't knock yourself out if you don't have time.

I was disappointed in the response at the other forums, but I can't say it's normal to get no responses. I posted at the Ubuntu forums which IMO are the most active and cooperative forums I have ever been on. Also I tried linuxquestions.org and linuxforums.org. Neither of them got any responses as well. I can only think that it is a pretty obscure and unique problem.

I can't say for sure that the RT mailing list would be a bad idea, but after trolling it for awhile I can tell I'd be out of my element, so I'd prolly make a nuisance of myself. Better for me to have a bit more knowledge before I tred in there. A few of those developers are bona fide legends, like Theodore T'so and Ingo Molnar. Guys like that are off the planet.

[Daniel ~]

Originally Posted by ccperf721p

That's for sure.. I haven't given up on putting my own package together. I've only had a couple of hours a week to work on it but it's getting close to at least booting to a command line.. I was unable to find a suitable base in another distribution so I decided to start from scratch. Maybe by the weekend I will have a kernel installed and can begin to test the base system.

You know me! If it's worth having...it's worth waiting for...":O}

[ccperf721p]

I loaded up and patched the kernel tonight just out of curiosity. I was able to produce a sympton similar to yours, except mine loads both cores at 50%.. I believe I have it narrowed down to a couple of options which I will know about after I finish compiling the kernel and reboot..


Edit: Well it was not what I was thinking it was. I didn't check to make sure everything was good before I tried this kernel so I am going to boot into my other kernel and make sure it is solid first.

I know have a thread in here somewhere where I was complaining about this very same issue on the 2.6.25 kernel, but it would do it without any patches..

Edit # 2: Loaded up the 2.6.30 gentoo-sources kernel and everything is fine, hitting all cores at 92 - 93%.. Not sure what is going on with the 2.6.29, but there is obviously some kind of conflict.. I do know the 2.6.30 is a stable kernel, and from what I can find there aren't any changes that are obvious..

[ThunderRd]

Yeah, the vanilla kernels with no patches run fine, all of the Zen patches I have tried work fine, the rt13 to rt16 patches over 2.6.26.8 work fine, but no other rt patches since then will load both cores on the client.

I wonder what

make defconfig

would do?

[ccperf721p]

I'd follow default config with a menuconfig just to make sure what you need gets loaded up.. Worth a shot..

I have tried what I could think of without any improvement..

Edit: What I think is happening is cpu scheduling going on a per user basis, 1/2 for user 1/2 for root available. As I can run the same thing as root and only get 50% but I can run one instance as root and one as user and get 98% load..

[ccperf721p]

Originally Posted by Daniel ~

You know me! If it's worth having...it's worth waiting for...":O}

The base system is pretty much much set, booted it for the first time about 20 minutes ago, basically all that means is that the easy part is over.

[ThunderRd]

Originally Posted by ccperf721p

I'd follow default config with a menuconfig just to make sure what you need gets loaded up.. Worth a shot..

I have tried what I could think of without any improvement..

Edit: What I think is happening is cpu scheduling going on a per user basis, 1/2 for user 1/2 for root available. As I can run the same thing as root and only get 50% but I can run one instance as root and one as user and get 98% load..

I try make defaultconfig later when I get home.

Re: the edit. I don't know much, but I doubt that is what is happening. Remember that the whole crux of the RT Kernel is that it is capable of preempting even parts of itself based on system demands.

Did you run the SMP client like this?

./fah6 -smp

You need the -smp flag or it runs in normal mode. Also,

./fah6 -smp -verbosity 9

makes the client more chatty and may reveal something.

[ccperf721p]

I had -smp set. Didn't use verb* settings though.

I couldn't sleep so I tried the same kernel and patch in the testing install on my main rig. Same results, even with -smp 8 (hyperthreading enabled)