HIGHMEM=4G slows down ps2pdf with 2.4.28
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 : 0vendor_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 KBfdiv_bug : nohlt_bug : nof00f_bug : nocoma_bug : no
fpu : yesfpu_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
Quoted 9 lines · Alain Tesio
Hi,
With a 2.4.28 kernel, 1.5 Go RAM and nothing exotic, everything works fine with
CONFIG_HIGHMEM4GandCONFIG_HIGHMEM=yexcept that
ps2pdf is about 30 times slowerps2pdf is a script using ghostscript with device=pdfwriter, I don't know if he's doing something special with memory allocations.
Quoted 8 lines
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.
Quoted 43 lines
#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 : 0vendor_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 KBfdiv_bug: nohlt_bug: nof00f_bug: nocoma_bug: no
fpu : yesfpu_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
Quoted 10 lines · On Thu, 2 Dec 2004 17:08:15 -0200 Marcelo Tosatti…
Quoted 5 lines · Alain Tesio
Hi,
With a 2.4.28 kernel, 1.5 Go RAM and nothing exotic, everything works fine with
CONFIG_HIGHMEM4GandCONFIG_HIGHMEM=yexcept that
ps2pdf is about 30 times slowerHow does
/proc/mtrrlook 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
Quoted 21 lines · Alain Tesio
Quoted 10 lines · On Thu, 2 Dec 2004 17:08:15 -0200 Marcelo Tosatti…
Quoted 5 lines · Alain Tesio
Hi,
With a 2.4.28 kernel, 1.5 Go RAM and nothing exotic, everything works fine with
CONFIG_HIGHMEM4GandCONFIG_HIGHMEM=yexcept that
ps2pdf is about 30 times slowerHow does
/proc/mtrrlook 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
Quoted 2 lines
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.
Quoted 25 lines · Marcelo Tosatti
Quoted 21 lines · Alain Tesio
Quoted 10 lines · On Thu, 2 Dec 2004 17:08:15 -0200 Marcelo Tosatti…
Quoted 5 lines · Alain Tesio
Hi,
With a 2.4.28 kernel, 1.5 Go RAM and nothing exotic, everything works fine with
CONFIG_HIGHMEM4GandCONFIG_HIGHMEM=yexcept that
ps2pdf is about 30 times slowerHow does
/proc/mtrrlook 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=1Quoted 2 lines
I don't think that the hosting company played with the bios settings, and I don't do anything special with memory.
Quoted 6 lines
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.