Gentoo Forums
Gentoo Forums
Gentoo Forums
Quick Search: in
thrashing with ck-sources r5
View unanswered posts
View posts from last 24 hours

Goto page Previous  1, 2  
Reply to topic    Gentoo Forums Forum Index Kernel & Hardware
View previous topic :: View next topic  
Author Message
lurid
Guru
Guru


Joined: 12 Mar 2003
Posts: 595
Location: Florida

PostPosted: Mon Apr 14, 2003 2:18 am    Post subject: Reply with quote

I downloaded the all-in-one patch.gz from the page and applied it to a vanilla kernel. I don't use XFS on any machine I own, so its never turned on in the kernel anyway. I can't say if this is the problem or not.. all I can say is that ckpre6 works very well.

Speaking of renicing X, this is right off Con's webpage:

Con Kolivas wrote:
NOTE: YOU MUST NOT RENICE X WITH THESE PATCHES. READ FAQ

---

Renicing X? Many distributions (eg Mandrake) start X by default at a nice of -10 to make it more responsive. This is a workaround for the old scheduler limitations and the new interactivity changes make this unecessary, and will actually decrease performance with this kernel.


I saw this today when downloading the ckpre6 patch. I admit that I was actually renicing X by using a wrapper that called X and automatically set its nice to -10. I removed this and reset everything to call X how it should be, then compiled the kernel. Everything has been peachy since.
Back to top
View user's profile Send private message
ferringb
Retired Dev
Retired Dev


Joined: 03 Apr 2003
Posts: 357

PostPosted: Mon Apr 14, 2003 5:03 am    Post subject: Re: pre6 query Reply with quote

puntium wrote:
lurid, ferringb, are you using XFS? In a response I got from Con, he said he suspected the XFS snapshot that he used in ck5

In my original playing around w/ the split patches, I was applying the xfs patch. why? It updated portions of the vfs. I honestly have no clue what it was doing, but I figured it was backported updates to the vfs code (finer spinlocks for instance).
puntium wrote:
Which patched kernel are you referring to? the split patches vs pre6 all in one? I ask because it seems like just changing a config parameter sounds like a bit of a hack or a "magic number".. if there is a bad interaction with some other part of the kernel, I guess I want to know before I go running it on my main machine with a big xfs data partition.

I started w/ a standard 2.4.20 kernel originally pulled from kernel.org- I would presume it's exactly the same as the vanilla-source, but being thorough... in the last 24 hours I've ran both a split patched kernel, and the pre6 patched kernel (currently running pre6). I think one thing to note, is that by the time I started in w/ the split patches the Makefile was claiming it was ck6 from the patches.

Basically, I think about the time I started on the split patches, whatever issue that ck5 had was resolved in the splits, cause I had none of the ck5 'all-in-one' patches problems.

As for changing a configure parameter (specifically killing preemption according to swat) to fix the problem, I agree. It seems as though preemption brought out some very nasty behavior with the ck5 patch. I was running w/ the same config for the split kernel and preck6- preemption enabled, low-latency enabled, O(1) enabled, ll via sysctl disabled, and xfs disabled (I don't use it). XFS's behaviour under the preck6 patch I can't tell you- I don't run XFS (somebody else care to answer that question?).

puntium wrote:
I thought the interactivity patch was the new stuff they came up with for 2.5.65 (i think it was). there was a post on slashdot or linuxtoday or something about it. Was a long mailing list discussion about whether renicing X was the right thing to do etc.


The interactivity patch was a backport of 65 features, off the top of my head (and not authorative), preemption, O(1) scheduler, low-latency, vm addons, tunable hz/scheduler and what is called desktop tuning. Note I haven't yet gone through to see what the desktop-tuning actually modifies- I assume it tweaks the default values of the scheduler/preemption/hz, but again, I haven't looked.

Going by ck's statements on his website, renicing X is a bad thing under the new patches, and since the last time I tried it was w/ my previous kernel patched with ac1, preempt, and O(1) kernel- it didn't seem to help, so I stopped renicing it.
puntium wrote:
Interesting that there's no difference that shows up in your diffs tho..

Note those diffs were a comparison between the split and preck6 at around 7-8am this morning... I did the diff to see what was different between my current split kernel (using split patches from around midnight) and the new preck6 to see if there was any major changes.

I did this mainly since my split kernel behaved fine, and to see if preck6 was likely to do the same- since then I've been using the preck6 without any issues.

It's actually behaved better in terms of memory, which I'm guessing is due to the removal of some updates/tweaks to nr_request in ll_rw_blk - my system now hovers w/ an expected/acceptable amount of free memory, rather then 4-20mb...
Back to top
View user's profile Send private message
puntium
n00b
n00b


Joined: 14 Mar 2003
Posts: 24

PostPosted: Wed Apr 16, 2003 1:00 am    Post subject: hmmm Reply with quote

Hmm, sounds a bit more encourging.. the reason I ask about XFS is that when I originally reported the problem to Con, he had said that his patches worked fine on his test machines, but that he had gotten reports from other XFS users that had the same problems I had seen.

But if you guys had seen similar problems w/o xfs (i mean actually mounting an xfs partition, not just applying the patch) then maybe it doesn't have to do with xfs at all.

Anyways, i guess i'll wait around till maybe someone writes an ebuild.. looks like on Con's page ck6 isn't "pre" anymore, but he also doesn't mention anything about interactivity.. (i wonder if its stil there or not) I guess I can live with the xfs-kernel for a while longer.. i just really want the interactivity stuff w/o jumping to 2.5.. with all these emerges going on the b/g its a pain to have to wait for them to finish to be able to use my machine for anything else..

Anyhow.. good luck, i'll report back if I eventually get stuff working..
Back to top
View user's profile Send private message
jaloha
n00b
n00b


Joined: 07 Apr 2003
Posts: 35
Location: Honolulu, Hawai'i

PostPosted: Thu Apr 17, 2003 4:50 am    Post subject: It is definately a memleak Reply with quote

It is causing a memory leak in the kernel. Originally I though it was associated with the NVidia driver, but I loaded the software one instead and the problem still exist. Has anyone tried ck-sources-2.4.20-r6 I'll install it this evening and post back to the forum if it fixes things.

UPDATE: I have been running the ck-source-2.4.20-r6 for about 12 hours now (I think) and it just freaked out again. The memory usage jumped from about @130 to 437. The good thing is that the system froze for about 5-10 seconds and then recovered (before it became to slow to use). Right now it is as responsive as normal, just the memory usage is really large. Also, as I open large programs, such as mozilla, they eat large amounts of RAM as usual 100 Megs or so, but the total memory usage does not increase. As I use the system the memory usage decreases (down to 399) as I type this. Maybe it is just not reporting it as free. If someone who knows the memory manager better than me can explain this, that would be great, but other than hanging for a few seconds, the new sources seem great.


Last edited by jaloha on Fri Apr 18, 2003 10:43 am; edited 1 time in total
Back to top
View user's profile Send private message
dizzogg
Tux's lil' helper
Tux's lil' helper


Joined: 03 May 2002
Posts: 131
Location: Virginia, USA

PostPosted: Thu Apr 17, 2003 5:32 am    Post subject: Reply with quote

I'm running ck6 at the moment and it is running much better than ck5. no problems yet.... :D
Back to top
View user's profile Send private message
Vancouverite
Apprentice
Apprentice


Joined: 28 Sep 2002
Posts: 162
Location: Vancouver, Canada

PostPosted: Thu Apr 17, 2003 5:58 am    Post subject: Reply with quote

I too have been running CK6 all day with no problems.
_________________
Screenshot
Back to top
View user's profile Send private message
puntium
n00b
n00b


Joined: 14 Mar 2003
Posts: 24

PostPosted: Thu Apr 17, 2003 7:25 am    Post subject: ck6 looks ok here too Reply with quote

Reporting back..

Built ck6 from the portage tree wiht 200hz preempt, low latency, and xfs.. xosview shows a more sane memory usage with my typical usage .. about 50% cache not 100% user+share.

and its nice and fast ;-)
Back to top
View user's profile Send private message
Edgaer
Tux's lil' helper
Tux's lil' helper


Joined: 01 Jul 2002
Posts: 145
Location: The Isle of the Sundered

PostPosted: Thu Apr 17, 2003 10:56 pm    Post subject: Reply with quote

I just upgraded to -ck6 and I'm experienceing the thrashing, then X goes funny and my box hardlocks and I've had to go back to -gentoo-r2, though I've experienced the thrashing with every kernel I've tried (gentoo, vanilla, ck4, and ck6). With the gentoo line being the most stable. With vanilla I'd even see kernel crashes after so much ram would get eaten up.

So far the two main culprits have been the blackdown JDK 1.3.whatever, some kde programs and screen savers, and big emerges. Things have gotten a little better with the new 1.4.1 JDK but I'm still having problems. Anybody got any suggestions (I'd like to be able to actually use the ck sources rather than booting it up and then reverting to the gentoo kernel when ckX is too unstable)
_________________
--First General Edgaer
Hope is the last to die. Unlike the icq flower.
Back to top
View user's profile Send private message
jaloha
n00b
n00b


Joined: 07 Apr 2003
Posts: 35
Location: Honolulu, Hawai'i

PostPosted: Thu Apr 17, 2003 11:18 pm    Post subject: Reply with quote

Quote:
I've experienced the thrashing with every kernel I've tried (gentoo, vanilla, ck4, and ck6). With the gentoo line being the most stable.


It has been my experience that the Gentoo kernel does the best job when the memory gets tight. However, that is not the problem we have been discussing here. It sounds like you have a different memory leak in a kernel driver (it if was in an application, you could run top, see which one and kill it and then the problem should go away). Do you have any unusual hardware or strange file systems mounted?
Back to top
View user's profile Send private message
puntium
n00b
n00b


Joined: 14 Mar 2003
Posts: 24

PostPosted: Tue Apr 22, 2003 11:25 pm    Post subject: interactivity update? Reply with quote

I just check Con Kolivas' page and he now has an extra patch to add the interactivity update to the O(1) scheduler. I guess this means that the ck6 patch does not include the update..

Has anyone that uses XFS tried the extra interactivity patch?
Back to top
View user's profile Send private message
Edgaer
Tux's lil' helper
Tux's lil' helper


Joined: 01 Jul 2002
Posts: 145
Location: The Isle of the Sundered

PostPosted: Mon Apr 28, 2003 4:19 am    Post subject: Reply with quote

jaloha wrote:
It has been my experience that the Gentoo kernel does the best job when the memory gets tight. However, that is not the problem we have been discussing here. It sounds like you have a different memory leak in a kernel driver (it if was in an application, you could run top, see which one and kill it and then the problem should go away). Do you have any unusual hardware or strange file systems mounted?


I don't know if it's a kernel issue or not. How would I go about finding out? And I shouldn't have anything weird going with the kernel. Ohter than the standard weirdness of having to compile scsi emulation in to burn cd's, the only file systems I keep open regularly are resiserfs, procfs, devfs, and tempfs. I frequently use vfat though I don't keep that mounted all the time anymore. And when I need at my /boot partition I add ext3 to the mix. And occasionally I'll have msdos or iso9660 file systems when I'm hunting through cd's and floppys.

And I noticed the thrashing start up today, so I fired up top and watched my free memory go from like 16M to 3-4M, and this was after closing a program that was eating up like 20% of my memory. Though that might have been conincindental. Though my system seems to have recovered this time as I'm typing this after shutting a few programs down. Though do you think this could be a video driver issue? I'm using the nvidia drivers though I may have also experienced this issue with the kernel's built in i810 driver as well (I don't recall if I did or not off hand)
_________________
--First General Edgaer
Hope is the last to die. Unlike the icq flower.
Back to top
View user's profile Send private message
jaloha
n00b
n00b


Joined: 07 Apr 2003
Posts: 35
Location: Honolulu, Hawai'i

PostPosted: Mon Apr 28, 2003 6:28 am    Post subject: Thrashing always Reply with quote

Edgaer,

It might just be that your system is very limited on memory. I don't know your system specifications or if the applications you use are memory intensive.

I would first enable the "nv" driver in your XF86Config file and confirm that the problem still exist.

Then, open top and look at the values in the VIRT and RES columns. I am not exceptionally knowledgeable about this type of thing, but I think RES is the amount of memory the application is using, and VIRT is the amount allocated to it. If the sum of the values in the RES is greater than the total amount of memory in your system less the size of you kernel memory usage, the system has to swap to disk. Most likely you will be able to find another application with excessive memory usage.

Also, make sure X is not reniced.

btw, how much memory do you have and what window manager are you running?
Back to top
View user's profile Send private message
Edgaer
Tux's lil' helper
Tux's lil' helper


Joined: 01 Jul 2002
Posts: 145
Location: The Isle of the Sundered

PostPosted: Wed May 14, 2003 11:46 pm    Post subject: Re: Thrashing always Reply with quote

jaloha wrote:
Edgaer,

It might just be that your system is very limited on memory. I don't know your system specifications or if the applications you use are memory intensive.


Actually I was playing the other day and noticed that rather than my memory being at issue I get an extremely large load average. It was like 5.xx, 3.xx, and I think either 1.xx or 2.xx

jaloha wrote:

I would first enable the "nv" driver in your XF86Config file and confirm that the problem still exist.

Then, open top and look at the values in the VIRT and RES columns. I am not exceptionally knowledgeable about this type of thing, but I think RES is the amount of memory the application is using, and VIRT is the amount allocated to it. If the sum of the values in the RES is greater than the total amount of memory in your system less the size of you kernel memory usage, the system has to swap to disk. Most likely you will be able to find another application with excessive memory usage.


I might try that. And where do I find out the kernel memory usage?

jaloha wrote:

Also, make sure X is not reniced.


Shouldn't be. Especially now that I've got the new gentoo sources with some of the ck patch stuff installed

jaloha wrote:

btw, how much memory do you have and what window manager are you running?


128Meg and kde or fluxbox depending on needs. Also, where can I get a sane layman's guide to which memory management patch tree to use?
_________________
--First General Edgaer
Hope is the last to die. Unlike the icq flower.
Back to top
View user's profile Send private message
jaloha
n00b
n00b


Joined: 07 Apr 2003
Posts: 35
Location: Honolulu, Hawai'i

PostPosted: Thu May 15, 2003 12:26 am    Post subject: Re: Thrashing always Reply with quote

Edgaer wrote:


And where do I find out the kernel memory usage?



I am not really the person to ask about this, but I think just run free and then subtract your system memory from the value in total.
EXAMPLE:
Code:

josh@aloha josh $ free
             total       used       free     shared    buffers     cached
Mem:        515004     507720       7284          0      33276     244208
-/+ buffers/cache:     230236     284768
Swap:       257032          0     257032


I have 512 Megs of RAM in my computer, so this should mean that the kernel is using

((512 * 1024) - 515004) / 1024 = (approx.) 9 Megs of memory.

Edgaer wrote:

128Meg and kde or fluxbox depending on needs. Also, where can I get a sane layman's guide to which memory management patch tree to use?


Under KDE with an average workload (which is not small for me) my system uses about 200-300 megs of memory. This is just what the applications require. Do you have gtop? This program give a good breakdown of current memory usage. The issue is that if your programs are using 200 Megs of RAM and the system only has 128, you will experience fairly intense swapping.

To answer you second question and further address this issue, I highly recommend the new 2.5.* mm sources (on the desktop). Configuring them and debugging problems with them can be more difficult, but they is a definate performance advantage. Use the CFQ scheduler. The default memory manager for this kernel seems really good for the desktop. A lot of people seem to enjoy it.
Back to top
View user's profile Send private message
Display posts from previous:   
Reply to topic    Gentoo Forums Forum Index Kernel & Hardware All times are GMT
Goto page Previous  1, 2
Page 2 of 2

 
Jump to:  
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum