[gentoo-user] Very bad system response

[gentoo-user] Very bad system response

[Michael Schreckenbauer]

Hi List,
I have a very strange thing here. I don't know when this started, but my 
feeling is, that it was around kernel 2.6.15 and it's getting worse with each 
version.
From time to time my system comes to a complete halt, that means not a single 
application shows any response to mouse and/or keyactions, sshing into it 
takes forever. After a while (up to ~30s ) normal reaction is back again. 
Strange thing is, there is no activity at the times this happens. Reading 
email and browsing the web is all that is needed. CPU is at 3-5% according to 
top, mem has ~1GB free according to free.
I can somehow force this behaviour by doing some harddisc io, eg untaring a 
kernel-tarball. But it does not happen everytime I do this.
This is really, really annoying. I have no idea, if this is a kernel, a FS or 
any other issue. Any help with this is most welcome.
Some additional info: fs is xfs, kernel is 2.6.19-gentoo-r4 and I'm a 
KDE-User, but this also happens with xfce4 ;)

Thanks,
Michael
--
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Very bad system response

[Hemmann, Volker Armin]

On Dienstag, 6. Februar 2007, Pierre-Yves Rofes wrote:
> On Tue, February 6, 2007 9:29 pm, Michael Schreckenbauer wrote:
> > Hi List,
> > I have a very strange thing here. I don't know when this started, but my
> > feeling is, that it was around kernel 2.6.15 and it's getting worse with
> > each
> > version.
> > From time to time my system comes to a complete halt, that means not a
> > single
> > application shows any response to mouse and/or keyactions, sshing into it
> > takes forever. After a while (up to ~30s ) normal reaction is back again.
> > Strange thing is, there is no activity at the times this happens. Reading
> > email and browsing the web is all that is needed. CPU is at 3-5%
> > according to
> > top, mem has ~1GB free according to free.
> > I can somehow force this behaviour by doing some harddisc io, eg untaring
> > a
> > kernel-tarball. But it does not happen everytime I do this.
> > This is really, really annoying. I have no idea, if this is a kernel, a
> > FS or
> > any other issue. Any help with this is most welcome.
> > Some additional info: fs is xfs, kernel is 2.6.19-gentoo-r4 and I'm a
> > KDE-User, but this also happens with xfce4 ;)
>
> Maybe a faulty hard drive with bad sectors that can't be read. Anything in
> /var/log/messages or any other log file?
>
> --
> Pierre-Yves Rofes

I am not sure that something is 'broken'. His problem is like mine... heavy 
disk i/o kills performance. And if the box hits swap, everything sucks.... 
because for some reason, always the wrong stuff is swapped out ;)

Really, sometimes, it is like every single bit is fetched individually from 
the swap space...
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Very bad system response

[Michael Schreckenbauer]

Hi,

Am Dienstag, 6. Februar 2007 schrieb Pierre-Yves Rofes:
> On Tue, February 6, 2007 9:29 pm, Michael Schreckenbauer wrote:
> > From time to time my system comes to a complete halt, that means not a
> > single
> > application shows any response to mouse and/or keyactions, sshing into it
> > takes forever. After a while (up to ~30s ) normal reaction is back again.
> > Strange thing is, there is no activity at the times this happens. Reading
> > email and browsing the web is all that is needed. CPU is at 3-5%
> > according to
> > top, mem has ~1GB free according to free.

> Maybe a faulty hard drive with bad sectors that can't be read. Anything in
> /var/log/messages or any other log file?

no sorry, I should have mentioned that. Nothing in the logs.
I am currently trying out, what Etaoin Shrdlu suggested in the 
thread "Performance problem at writing big files and Multitasking".
I switched from cfq to deadline scheduler. No hang since I did that. Maybe 
that did the trick? I'll let you know :)

> Pierre-Yves Rofes

Thanks,
Michael
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Very bad system response

[Michael Schreckenbauer]

Hi,

Am Dienstag, 6. Februar 2007 schrieb Hemmann, Volker Armin:
> On Dienstag, 6. Februar 2007, Pierre-Yves Rofes wrote:
> > On Tue, February 6, 2007 9:29 pm, Michael Schreckenbauer wrote:
> > > From time to time my system comes to a complete halt, that means not a
> > > single
> > > application shows any response to mouse and/or keyactions, sshing into
> > > it takes forever. After a while (up to ~30s ) normal reaction is back
> > > again. Strange thing is, there is no activity at the times this
> > > happens. Reading email and browsing the web is all that is needed. CPU
> > > is at 3-5% according to
> > > top, mem has ~1GB free according to free.
> > Maybe a faulty hard drive with bad sectors that can't be read. Anything
> > in /var/log/messages or any other log file?
> >
> > --
> > Pierre-Yves Rofes
>
> I am not sure that something is 'broken'. His problem is like mine... heavy
> disk i/o kills performance. And if the box hits swap, everything sucks....
> because for some reason, always the wrong stuff is swapped out ;)
> Really, sometimes, it is like every single bit is fetched individually from
> the swap space...

I don't think, swap is the cause of this. This system has 1,5GB Ram, swap is 
nearly never touched :)

Thanks,
Michael
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Very bad system response

[Hemmann, Volker Armin]

On Dienstag, 6. Februar 2007, Michael Schreckenbauer wrote:
> Hi,
>
> Am Dienstag, 6. Februar 2007 schrieb Hemmann, Volker Armin:
> > On Dienstag, 6. Februar 2007, Pierre-Yves Rofes wrote:
> > > On Tue, February 6, 2007 9:29 pm, Michael Schreckenbauer wrote:
> > > > From time to time my system comes to a complete halt, that means not
> > > > a single
> > > > application shows any response to mouse and/or keyactions, sshing
> > > > into it takes forever. After a while (up to ~30s ) normal reaction is
> > > > back again. Strange thing is, there is no activity at the times this
> > > > happens. Reading email and browsing the web is all that is needed.
> > > > CPU is at 3-5% according to
> > > > top, mem has ~1GB free according to free.
> > >
> > > Maybe a faulty hard drive with bad sectors that can't be read. Anything
> > > in /var/log/messages or any other log file?
> > >
> > > --
> > > Pierre-Yves Rofes
> >
> > I am not sure that something is 'broken'. His problem is like mine...
> > heavy disk i/o kills performance. And if the box hits swap, everything
> > sucks.... because for some reason, always the wrong stuff is swapped out
> > ;) Really, sometimes, it is like every single bit is fetched individually
> > from the swap space...
>
> I don't think, swap is the cause of this. This system has 1,5GB Ram, swap
> is nearly never touched :)
>
>

swap was only an example. Things that kill performance: big tar, big copy 
around, and everything that touches a lot of files....
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Very bad system response

[Norberto Bensa]

[-- Attachment #1: Type: text/plain, Size: 380 bytes --]

Hemmann, Volker Armin wrote:
> On Dienstag, 6. Februar 2007, Michael Schreckenbauer wrote:
> > I don't think, swap is the cause of this. This system has 1,5GB Ram, swap
> > is nearly never touched :)
>
> swap was only an example. Things that kill performance: big tar, big copy
> around, and everything that touches a lot of files....

Check your DMA settings


Regards,
Norberto

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-user] Very bad system response

[Hemmann, Volker Armin]

On Mittwoch, 7. Februar 2007, Norberto Bensa wrote:
> Hemmann, Volker Armin wrote:
> > On Dienstag, 6. Februar 2007, Michael Schreckenbauer wrote:
> > > I don't think, swap is the cause of this. This system has 1,5GB Ram,
> > > swap is nearly never touched :)
> >
> > swap was only an example. Things that kill performance: big tar, big copy
> > around, and everything that touches a lot of files....
>
> Check your DMA settings
>

my DMA settings are ok. Really, really.
hdparm /dev/hda

/dev/hda:
 multcount    = 16 (on)
 IO_support   =  1 (32-bit)
 unmaskirq    =  1 (on)
 using_dma    =  1 (on)
 keepsettings =  0 (off)
 readonly     =  0 (off)
 readahead    = 256 (on)
 geometry     = 16383/255/63, sectors = 234493056, start = 0

[   36.419597] hda: 234493056 sectors (120060 MB) w/8192KiB Cache, 
CHS=16383/255/63, UDMA(100)

and

[   37.500706] ahci 0000:03:00.0: AHCI 0001.0000 32 slots 1 ports 3 Gbps 0x1 
impl SATA mode
[   37.501239] ahci 0000:03:00.0: flags: 64bit ncq pm led clo pmp pio slum 
part
[   37.501632] ata1: SATA max UDMA/133 cmd 0xFFFFC20000004100 ctl 0x0 bmdma 
0x0 irq 35
[   37.502167] scsi0 : ahci
[   37.987972] ata1: SATA link up 3.0 Gbps (SStatus 123 SControl 300)
[   37.988743] ata1.00: ATA-7, max UDMA/133, 312581808 sectors: LBA48
[   37.989026] ata1.00: ata1: dev 0 multi count 16
[   37.989838] ata1.00: configured for UDMA/133
[   37.990176] scsi 0:0:0:0: Direct-Access     ATA      WDC WD1600JS-00M 10.0 
PQ: 0 ANSI: 5
[   37.990773] SCSI device sda: 312581808 512-byte hdwr sectors (160042 MB)
[   37.991061] sda: Write Protect is off
[   37.991339] sda: Mode Sense: 00 3a 00 00
[   37.991349] SCSI device sda: drive cache: write back
[   37.991657] SCSI device sda: 312581808 512-byte hdwr sectors (160042 MB)
[   37.991942] sda: Write Protect is off
[   37.992225] sda: Mode Sense: 00 3a 00 00
[   37.992235] SCSI device sda: drive cache: write back
[   37.992514]  sda: sda1 sda2 sda3 sda4 < sda5 sda6 >
[   38.033040] sd 0:0:0:0: Attached scsi disk sda

-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Very bad system response

[Michael Schreckenbauer]

Hi,

Am Mittwoch, 7. Februar 2007 04:49 schrieb Norberto Bensa:
> Hemmann, Volker Armin wrote:
> > On Dienstag, 6. Februar 2007, Michael Schreckenbauer wrote:
> > > I don't think, swap is the cause of this. This system has 1,5GB Ram,
> > > swap is nearly never touched :)
> >
> > swap was only an example. Things that kill performance: big tar, big copy
> > around, and everything that touches a lot of files....

> Check your DMA settings

DMA is ok. As I wrote in my original report, the hang lasts max 30s, then 
everything is ok again regardless of io activity. The io-scheduler thingy 
another member suggested in another thread works so far :)

> Regards,
> Norberto

Thanks,
Michael
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Very bad system response

[Benno Schulenberg]

Hemmann, Volker Armin wrote:
> hdparm /dev/hda
>
> /dev/hda:
>  multcount    = 16 (on)
>  IO_support   =  1 (32-bit)
>  unmaskirq    =  1 (on)
>  using_dma    =  1 (on)
>  keepsettings =  0 (off)
>  readonly     =  0 (off)
>  readahead    = 256 (on)
>  geometry     = 16383/255/63, sectors = 234493056, start = 0

Maybe set readahead to a smaller value?  Something like 8 or 16?

At least that's what http://linuxgazette.net/issue79/punk.html 
advises.  If you have specific reasons for the higher value, then 
please elaborate.

Benno
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Very bad system response

[Michael Schreckenbauer]

Hi,

Am Mittwoch, 7. Februar 2007 12:35 schrieb Benno Schulenberg:
> Hemmann, Volker Armin wrote:
> > hdparm /dev/hda
> >
> > /dev/hda:
> >  multcount    = 16 (on)
> >  IO_support   =  1 (32-bit)
> >  unmaskirq    =  1 (on)
> >  using_dma    =  1 (on)
> >  keepsettings =  0 (off)
> >  readonly     =  0 (off)
> >  readahead    = 256 (on)
> >  geometry     = 16383/255/63, sectors = 234493056, start = 0
>
> Maybe set readahead to a smaller value?  Something like 8 or 16?
>
> At least that's what http://linuxgazette.net/issue79/punk.html
> advises.  If you have specific reasons for the higher value, then
> please elaborate.

I believe there is a misunderstanding here. I am the one with the problem not 
Volker :)
Unfortunately I'm not in front of the affected machine right now, but I'm 
quite shure I use the defaults as I never tuned the discs with hdparm

> Benno

Thanks,
Michael
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Very bad system response

[Hemmann, Volker Armin]

On Mittwoch, 7. Februar 2007, Michael Schreckenbauer wrote:
> Hi,
>
> Am Mittwoch, 7. Februar 2007 12:35 schrieb Benno Schulenberg:
> > Hemmann, Volker Armin wrote:
> > > hdparm /dev/hda
> > >
> > > /dev/hda:
> > >  multcount    = 16 (on)
> > >  IO_support   =  1 (32-bit)
> > >  unmaskirq    =  1 (on)
> > >  using_dma    =  1 (on)
> > >  keepsettings =  0 (off)
> > >  readonly     =  0 (off)
> > >  readahead    = 256 (on)
> > >  geometry     = 16383/255/63, sectors = 234493056, start = 0
> >
> > Maybe set readahead to a smaller value?  Something like 8 or 16?
> >
> > At least that's what http://linuxgazette.net/issue79/punk.html
> > advises.  If you have specific reasons for the higher value, then
> > please elaborate.

it is a default setting I never touched?
Aside from 'IO_support' everything is default.

Oh, and I tried different io-scheds without any success.

>
> I believe there is a misunderstanding here. I am the one with the problem
> not Volker :)

we have both similar problems ;)
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Very bad system response

[Michael Schreckenbauer]

Hi,

Am Mittwoch, 7. Februar 2007 21:19 schrieb Hemmann, Volker Armin:
> On Mittwoch, 7. Februar 2007, Michael Schreckenbauer wrote:
> > Hi,
> >
> > Am Mittwoch, 7. Februar 2007 12:35 schrieb Benno Schulenberg:
> > > Hemmann, Volker Armin wrote:
> > > > hdparm /dev/hda
> > > >
> > > > /dev/hda:
> > > >  multcount    = 16 (on)
> > > >  IO_support   =  1 (32-bit)
> > > >  unmaskirq    =  1 (on)
> > > >  using_dma    =  1 (on)
> > > >  keepsettings =  0 (off)
> > > >  readonly     =  0 (off)
> > > >  readahead    = 256 (on)
> > > >  geometry     = 16383/255/63, sectors = 234493056, start = 0
> > >
> > > Maybe set readahead to a smaller value?  Something like 8 or 16?
> > >
> > > At least that's what http://linuxgazette.net/issue79/punk.html
> > > advises.  If you have specific reasons for the higher value, then
> > > please elaborate.
>
> it is a default setting I never touched?
> Aside from 'IO_support' everything is default.

same here.

> Oh, and I tried different io-scheds without any success.

anticipatory and deadline work better for me, but far from perfect. cfq is 
unusable.

> > I believe there is a misunderstanding here. I am the one with the problem
> > not Volker :)
> we have both similar problems ;)

Ah! Nice to hear I am not alone! *g*
googling around, I found some people having a similar problem too, all of them 
are using xfs as I do (eg http://www.thisishull.net/showthread.php?t=219580). 
What's your fs? Mounting with nobarrier didn't work for me :(

Regards,
Michael
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Very bad system response

[Hemmann, Volker Armin]

On Donnerstag, 8. Februar 2007, Michael Schreckenbauer wrote:
> Hi,
>
> Am Mittwoch, 7. Februar 2007 21:19 schrieb Hemmann, Volker Armin:
> > On Mittwoch, 7. Februar 2007, Michael Schreckenbauer wrote:
> > > Hi,
> > >
> > > Am Mittwoch, 7. Februar 2007 12:35 schrieb Benno Schulenberg:
> > > > Hemmann, Volker Armin wrote:
> > > > > hdparm /dev/hda
> > > > >
> > > > > /dev/hda:
> > > > >  multcount    = 16 (on)
> > > > >  IO_support   =  1 (32-bit)
> > > > >  unmaskirq    =  1 (on)
> > > > >  using_dma    =  1 (on)
> > > > >  keepsettings =  0 (off)
> > > > >  readonly     =  0 (off)
> > > > >  readahead    = 256 (on)
> > > > >  geometry     = 16383/255/63, sectors = 234493056, start = 0
> > > >
> > > > Maybe set readahead to a smaller value?  Something like 8 or 16?
> > > >
> > > > At least that's what http://linuxgazette.net/issue79/punk.html
> > > > advises.  If you have specific reasons for the higher value, then
> > > > please elaborate.
> >
> > it is a default setting I never touched?
> > Aside from 'IO_support' everything is default.
>
> same here.
>
> > Oh, and I tried different io-scheds without any success.
>
> anticipatory and deadline work better for me, but far from perfect. cfq is
> unusable.

as was simply horrible for me. Much worse than cfq. But cfq is bad too... and 
deadline is not much better.

>
> > > I believe there is a misunderstanding here. I am the one with the
> > > problem not Volker :)
> >
> > we have both similar problems ;)
>
> Ah! Nice to hear I am not alone! *g*
> googling around, I found some people having a similar problem too, all of
> them are using xfs as I do (eg
> http://www.thisishull.net/showthread.php?t=219580). What's your fs?

reiserfs
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Very bad system response

[Robert Szentmihalyi]

On Tuesday 06 February 2007 21:29, Michael Schreckenbauer wrote:
> Hi List,

Hi Michael,

Good to see yoou on the Gentoo list! :-)

> I have a very strange thing here. I don't know when this started, but my
> feeling is, that it was around kernel 2.6.15 and it's getting worse with
> each version.
> From time to time my system comes to a complete halt, that means not a
> single application shows any response to mouse and/or keyactions, sshing
> into it takes forever. After a while (up to ~30s ) normal reaction is back
> again. Strange thing is, there is no activity at the times this happens.
> Reading email and browsing the web is all that is needed. CPU is at 3-5%
> according to top, mem has ~1GB free according to free.
> I can somehow force this behaviour by doing some harddisc io, eg untaring a
> kernel-tarball. But it does not happen everytime I do this.
> This is really, really annoying. I have no idea, if this is a kernel, a FS
> or any other issue. Any help with this is most welcome.
> Some additional info: fs is xfs, kernel is 2.6.19-gentoo-r4 and I'm a
> KDE-User, but this also happens with xfce4 ;)

I had this problem back with older 2.6 kernels and the problem turned out to 
be caused by XFS.
XFS is not a good choice for a desktop, as for my experience.
I bet the problem will be gone once you switch to ReiserFS or ext3.



>
> Thanks,
> Michael

Cheers,
 Robert
-- 
gentoo-user@gentoo.org mailing list