CPU usage problem with custom 2.6.7 kernel

CPU usage problem with custom 2.6.7 kernel

[Patrick Coleman]

Hi,
I am using reiser3 on linux-2.6.7 on both a desktop and a fileserver. Intensive reading and writing
seem to be using 100% cpu on both, limiting the rate at which I can transfer stuff to/from the
fileserver to about 1Mb/s (on a 100MB/s network/800mhz Celeron server). The default slackware-2.4
kernel does not seem to have this problem (the slackware 2.6.7 kernel does not have reiserfs support
by default). Anyone have any ideas?

Thanks in advance,
Patrick

hdparm output (desktop) - 80gb (seagate), hda3 is 65gb reiser3:
pcoleman@hal:~$ sudo hdparm /dev/hda

/dev/hda:
   multcount    =  0 (off)
   IO_support   =  1 (32-bit)
   unmaskirq    =  1 (on)
   using_dma    =  1 (on)
   keepsettings =  0 (off)
   readonly     =  0 (off)
   readahead    = 256 (on)
   geometry     = 65535/16/63, sectors = 156301488, start = 0
pcoleman@hal:~$

hdparm output (server) - 120gb (western digital), one partition, hdb1 is 120gb reiser3:
admin@server:~$ sudo hdparm /dev/hda

/dev/hda:
   multcount    = 16 (on)
   IO_support   =  0 (default 16-bit)
   unmaskirq    =  0 (off)
   using_dma    =  0 (off)
   keepsettings =  0 (off)
   readonly     =  0 (off)
   readahead    = 256 (on)
   geometry     = 16383/255/63, sectors = 234441648, start = 0
admin@server:~$

I can supply kernel config files if required (the mailing list rejects messages over 40kb)

Re: CPU usage problem with custom 2.6.7 kernel

[Spam]

> Hi,
> I am using reiser3 on linux-2.6.7 on both a desktop and a
> fileserver. Intensive reading and writing
> seem to be using 100% cpu on both, limiting the rate at which I can transfer stuff to/from the
> fileserver to about 1Mb/s (on a 100MB/s network/800mhz Celeron
> server). The default slackware-2.4
> kernel does not seem to have this problem (the slackware 2.6.7
> kernel does not have reiserfs support
> by default). Anyone have any ideas?

> Thanks in advance,
> Patrick

> hdparm output (desktop) - 80gb (seagate), hda3 is 65gb reiser3:
> pcoleman@hal:~$ sudo hdparm /dev/hda

> /dev/hda:
>    multcount    =  0 (off)
>    IO_support   =  1 (32-bit)
>    unmaskirq    =  1 (on)
>    using_dma    =  1 (on)
>    keepsettings =  0 (off)
>    readonly     =  0 (off)
>    readahead    = 256 (on)
>    geometry     = 65535/16/63, sectors = 156301488, start = 0
> pcoleman@hal:~$

You really should enable multicount by using: hdparm -m16 /dev/hda

> hdparm output (server) - 120gb (western digital), one partition, hdb1 is 120gb reiser3:
> admin@server:~$ sudo hdparm /dev/hda

> /dev/hda:
>    multcount    = 16 (on)
>    IO_support   =  0 (default 16-bit)
>    unmaskirq    =  0 (off)
>    using_dma    =  0 (off)
>    keepsettings =  0 (off)
>    readonly     =  0 (off)
>    readahead    = 256 (on)
>    geometry     = 16383/255/63, sectors = 234441648, start = 0
> admin@server:~$

> I can supply kernel config files if required (the mailing list rejects messages over 40kb)


You need to enable DMA and 32bit I/O!
Use: hdparm -d1c3 /dev/hda

Other options to increas performance a lot are drive read-lookahead and write
cache: hdparm -A1W1 /dev/hda
Just a warning with write cache. If you loose power then you may loose
data (even though journals say otherwise).

You may also want to use unmask IRQ to allow other hardware like
network cards, to use interrupts while disk is reading/writing.
Use: hdparm -u1 /dev/hda

Re: CPU usage problem with custom 2.6.7 kernel

[Chris Mason]

On Sat, 2004-08-14 at 06:36, Spam wrote:
> > 
> You need to enable DMA and 32bit I/O!
> Use: hdparm -d1c3 /dev/hda
> 

DMA is the major problem there.  hdparm -d1 /dev/hda should be enough to
fix the performance problems.  The rest of the hints given in the
message are just tuning on top of dma, I'd suggest reading the hdparm
manpage carefully before changing any of them.

-chris