HIGHMEM=4G slows down ps2pdf with 2.4.28

HIGHMEM=4G slows down ps2pdf with 2.4.28

[Alain Tesio]

Hi,

With a 2.4.28 kernel, 1.5 Go RAM and nothing exotic, everything works fine
with CONFIG_HIGHMEM4G and CONFIG_HIGHMEM=y except that
ps2pdf is about 30 times slower 

ps2pdf is a script using ghostscript with device=pdfwriter, I don't know if he's
doing something special with memory allocations.

If you want to test it, use this file:
http://www.floc.net/observer/USDP/hoyteclassical/hoyteclassical.ps
(gs-gpl 8.01 on debian sid)

Other usual server daemons seems unaffected.

If this is a known behaviour for HIGHMEM to slow down random apps, it will
be nice to put a warning in the make config help !


Alain




#lspci | grep bridge
cat /proc0000:00:00.0 Host bridge: Intel Corp. 82865G/PE/P DRAM Controller/Host-Hub Interface (rev 02)
0000:00:1e.0 PCI bridge: Intel Corp. 82801 PCI Bridge (rev c2)
0000:00:1f.0 ISA bridge: Intel Corp. 82801EB/ER (ICH5/ICH5R) LPC Bridge (rev 02)

#cat /proc/meminfo
        total:    used:    free:  shared: buffers:  cached:
Mem:  1579692032 1522913280 56778752        0 194080768 798294016
Swap: 2147467264 19042304 2128424960
MemTotal:      1542668 kB
MemFree:         55448 kB
MemShared:           0 kB
Buffers:        189532 kB
Cached:         766428 kB
SwapCached:      13156 kB
Active:         461848 kB
Inactive:       800564 kB
HighTotal:      646336 kB
HighFree:         2044 kB
LowTotal:       896332 kB
LowFree:         53404 kB
SwapTotal:     2097136 kB
SwapFree:      2078540 kB

#cat /proc/cpuinfo
processor       : 0
vendor_id       : GenuineIntel
cpu family      : 15
model           : 2
model name      : Intel(R) Pentium(R) 4 CPU 3.40GHz
stepping        : 9
cpu MHz         : 3400.211
cache size      : 512 KB
fdiv_bug        : no
hlt_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 2
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe cid
bogomips        : 6789.52

# sh /usr/src/linux-2.4.25/scripts/ver_linux
If some fields are empty or look unusual you may have an old version.
Compare to the current minimal requirements in Documentation/Changes.

Linux onesite 2.4.28 #2 Tue Nov 30 16:28:30 MST 2004 i686 GNU/Linux

Gnu C                  3.3.4
Gnu make               3.80
util-linux             2.12
mount                  2.12
modutils               2.4.26
e2fsprogs              1.27
quota-tools            3.04.
Linux C Library        2.3.2
Dynamic linker (ldd)   2.3.2
Procps                 3.2.1
Net-tools              1.60
Console-tools          0.2.3
Sh-utils               5.0
Modules Loaded         af_packet

$time ps2pdf test.ps test.pdf

real    1m5.554s
user    0m17.470s
sys     0m0.040s

# now without highmem=4GO (lower load however)
$time ps2pdf test.ps test.pdf

real    0m0.667s
user    0m0.670s
sys     0m0.000s

Re: HIGHMEM=4G slows down ps2pdf with 2.4.28

[Marcelo Tosatti]

On Wed, Dec 01, 2004 at 11:25:22PM +0100, Alain Tesio wrote:
> Hi,
> 
> With a 2.4.28 kernel, 1.5 Go RAM and nothing exotic, everything works fine
> with CONFIG_HIGHMEM4G and CONFIG_HIGHMEM=y except that
> ps2pdf is about 30 times slower 
> 
> ps2pdf is a script using ghostscript with device=pdfwriter, I don't know if he's
> doing something special with memory allocations.
> 
> If you want to test it, use this file:
> http://www.floc.net/observer/USDP/hoyteclassical/hoyteclassical.ps
> (gs-gpl 8.01 on debian sid)
> 
> Other usual server daemons seems unaffected.
> 
> If this is a known behaviour for HIGHMEM to slow down random apps, it will
> be nice to put a warning in the make config help !

How does /proc/mtrr look like? 

Maybe some of your memory is configured as uncacheable.

> #lspci | grep bridge
> cat /proc0000:00:00.0 Host bridge: Intel Corp. 82865G/PE/P DRAM Controller/Host-Hub Interface (rev 02)
> 0000:00:1e.0 PCI bridge: Intel Corp. 82801 PCI Bridge (rev c2)
> 0000:00:1f.0 ISA bridge: Intel Corp. 82801EB/ER (ICH5/ICH5R) LPC Bridge (rev 02)
> 
> #cat /proc/meminfo
>         total:    used:    free:  shared: buffers:  cached:
> Mem:  1579692032 1522913280 56778752        0 194080768 798294016
> Swap: 2147467264 19042304 2128424960
> MemTotal:      1542668 kB
> MemFree:         55448 kB
> MemShared:           0 kB
> Buffers:        189532 kB
> Cached:         766428 kB
> SwapCached:      13156 kB
> Active:         461848 kB
> Inactive:       800564 kB
> HighTotal:      646336 kB
> HighFree:         2044 kB
> LowTotal:       896332 kB
> LowFree:         53404 kB
> SwapTotal:     2097136 kB
> SwapFree:      2078540 kB
> 
> #cat /proc/cpuinfo
> processor       : 0
> vendor_id       : GenuineIntel
> cpu family      : 15
> model           : 2
> model name      : Intel(R) Pentium(R) 4 CPU 3.40GHz
> stepping        : 9
> cpu MHz         : 3400.211
> cache size      : 512 KB
> fdiv_bug        : no
> hlt_bug         : no
> f00f_bug        : no
> coma_bug        : no
> fpu             : yes
> fpu_exception   : yes
> cpuid level     : 2
> wp              : yes
> flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe cid
> bogomips        : 6789.52

Re: HIGHMEM=4G slows down ps2pdf with 2.4.28

[Alain Tesio]

On Thu, 2 Dec 2004 17:08:15 -0200
Marcelo Tosatti <marcelo.tosatti@cyclades.com> wrote:

> On Wed, Dec 01, 2004 at 11:25:22PM +0100, Alain Tesio wrote:
> > Hi,
> > 
> > With a 2.4.28 kernel, 1.5 Go RAM and nothing exotic, everything works fine
> > with CONFIG_HIGHMEM4G and CONFIG_HIGHMEM=y except that
> > ps2pdf is about 30 times slower 

> How does /proc/mtrr look like? 
> 
> Maybe some of your memory is configured as uncacheable.

reg00: base=0x00000000 (   0MB), size=1024MB: write-back, count=1
reg01: base=0x40000000 (1024MB), size= 256MB: write-back, count=1
reg02: base=0x50000000 (1280MB), size= 128MB: write-back, count=1
reg03: base=0x58000000 (1408MB), size=  64MB: write-back, count=1
reg04: base=0x5c000000 (1472MB), size=  32MB: write-back, count=1
reg05: base=0x5e000000 (1504MB), size=  16MB: write-back, count=1

I don't think that the hosting company played with the bios settings,
and I don't do anything special with memory.

Alain

Re: HIGHMEM=4G slows down ps2pdf with 2.4.28

[Marcelo Tosatti]

On Fri, Dec 03, 2004 at 09:58:19PM +0100, Alain Tesio wrote:
> On Thu, 2 Dec 2004 17:08:15 -0200
> Marcelo Tosatti <marcelo.tosatti@cyclades.com> wrote:
> 
> > On Wed, Dec 01, 2004 at 11:25:22PM +0100, Alain Tesio wrote:
> > > Hi,
> > > 
> > > With a 2.4.28 kernel, 1.5 Go RAM and nothing exotic, everything works fine
> > > with CONFIG_HIGHMEM4G and CONFIG_HIGHMEM=y except that
> > > ps2pdf is about 30 times slower 
> 
> > How does /proc/mtrr look like? 
> > 
> > Maybe some of your memory is configured as uncacheable.
> 
> reg00: base=0x00000000 (   0MB), size=1024MB: write-back, count=1
> reg01: base=0x40000000 (1024MB), size= 256MB: write-back, count=1
> reg02: base=0x50000000 (1280MB), size= 128MB: write-back, count=1
> reg03: base=0x58000000 (1408MB), size=  64MB: write-back, count=1
> reg04: base=0x5c000000 (1472MB), size=  32MB: write-back, count=1
> reg05: base=0x5e000000 (1504MB), size=  16MB: write-back, count=1
> 
> I don't think that the hosting company played with the bios settings,
> and I don't do anything special with memory.

Alain,

All memory is correctly configured in MTRR it seems (all of it is write-back cacheable).

Do you have CONFIG_HIGHIO=y ? That might help a lot. The kernel has to use 
bounce buffers for IO otherwise.

Re: HIGHMEM=4G slows down ps2pdf with 2.4.28

[Marcelo Tosatti]

On Mon, Dec 06, 2004 at 04:41:18PM -0200, Marcelo Tosatti wrote:
> On Fri, Dec 03, 2004 at 09:58:19PM +0100, Alain Tesio wrote:
> > On Thu, 2 Dec 2004 17:08:15 -0200
> > Marcelo Tosatti <marcelo.tosatti@cyclades.com> wrote:
> > 
> > > On Wed, Dec 01, 2004 at 11:25:22PM +0100, Alain Tesio wrote:
> > > > Hi,
> > > > 
> > > > With a 2.4.28 kernel, 1.5 Go RAM and nothing exotic, everything works fine
> > > > with CONFIG_HIGHMEM4G and CONFIG_HIGHMEM=y except that
> > > > ps2pdf is about 30 times slower 
> > 
> > > How does /proc/mtrr look like? 
> > > 
> > > Maybe some of your memory is configured as uncacheable.
> > 
> > reg00: base=0x00000000 (   0MB), size=1024MB: write-back, count=1
> > reg01: base=0x40000000 (1024MB), size= 256MB: write-back, count=1
> > reg02: base=0x50000000 (1280MB), size= 128MB: write-back, count=1
> > reg03: base=0x58000000 (1408MB), size=  64MB: write-back, count=1
> > reg04: base=0x5c000000 (1472MB), size=  32MB: write-back, count=1
> > reg05: base=0x5e000000 (1504MB), size=  16MB: write-back, count=1
> > 
> > I don't think that the hosting company played with the bios settings,
> > and I don't do anything special with memory.
> 
> Alain,
> 
> All memory is correctly configured in MTRR it seems (all of it is write-back cacheable).
> 
> Do you have CONFIG_HIGHIO=y ? That might help a lot. The kernel has to use 
> bounce buffers for IO otherwise.

However you report 30x slower without HIGHMEM - thats a bit too much for 
the added overhead of bounce buffers. Anyway, try CONFIG_HIGHIO=y if you 
dont have it configured this way.