Skip to contentSkip to conversation

2.5.64-mm6

DISCUSSION dada1 1 participant 1 message Mar 13, 2003

lore
Share & tools
lore

dada1

author
#0 lore raw
Quoted 7 lines

This means that when an executable is first mapped in, the kernel will slurp the whole thing off disk in one hit. Some IO changes were made to speed this up.

This means that large cache-cold executables start significantly faster. Launching X11+KDE+mozilla goes from 23 seconds to 16. Starting

OpenOffice

Quoted 2 lines

seems to be 2x to 3x faster, and starting Konqueror maybe 3x faster too. Interesting.

Sounds nice indeed.

Do you think we could try to use 4Mo pages instead of 4Ko to map some libraries/executables ?

libc.so could be linked to align text to a 4Mo boundary, and to set a an ELF marker to tell the kernel/loader to try to use
HugeTLB page to map the text portion. The space wasted (the current glibc use about 2Mo instead of 4Mo) is neglictable if the same page is used for all processes in the machine...

HPUX11 has a chattr command that can mark binaries to use BigPages (for text pages or data pages). We could 'mark' big programs with an ELF attribute to ask for the behavior you describe (force the load of all the mapped portion at mmap time) or/and to try to use HugeTLBPages...

Eric