mmap()/mlockall()/read()
Benno Senoner discourseth:
Quoted 5 lines
Eric,
when not mlock()-ing the entire process (or at least the RT relevant data), you have no chance to give ANY guarantee thet some pages could not be swapped out.
That's exactly my concern..I have no guarantees that I won't run into N page-faults in a given period and miss my deadlines. Unfortunately, locking and mapping aren't that flexible. I need to use mlockall(MCL_CURRENT | MCL_FUTURE) to make sure I have no page faults. Then anything I mmap() will be initially locked as well..even though it may be larger than my RAM. A MAP_UNLOCKED option to mmap() would be a non-portable solution.
Quoted 3 lines
What I'm trying to say:
read()doesn't work better thanmmap()in terms of behaviour during swapping, because the kernel could easily swap out your buffer where youread()in the data.
Actually, the last time I tested this extensively (with a 2.0.x kernel I think), mmap() was much better performing than read()..for a while. When sequential accessing of a file got to a point around my RAM limits, the disk activity became insane. Maybe things are better now.
Quoted 4 lines
I will add a feature to the pagefaulter thread which uses mlock()/munlock if root privileges are available, so that even
Eric is satisfied.
:-)
Eric is hard to satisfy. :)
Quoted 4 lines
PS:
mlockall(MCL_CURRENT|MCL_FUTURE)is a bad idea because when you do themmap()of the large file the process tries to load all into mem,and my scheme would not work. Better to usemlockall(MCL_CURRENT)and mlock()/munlock() areas on demand.
This is a useful solution in some circumstances. However, it means I can't use malloc()/new, may have trouble with shared libraries or dynamically loaded plugins and have to worry about reserving stack space. It's not the way I want to work.
Eric
I agree no all topics,
using mlockall(MCL_CURRENT|MCL_FUTURE) on a low mem box, and trying to mmap a large file would simply not work.
I would definitively use MAP_UNLOCKED on Linux, becaue it could give us the best possible performance. (and maybe provide a fallback solution for other OSes)
I think the best fallback solution is the following (but not 100% reliable in the case of syscalls):mlock(MCL_CURRENT);mlock() malloec()ed aread
( or is there a better way to do this ?)
but strange,MAP_UNLOCKED seems to not exist on my 2.2.12 kernel !! I grepped through the entire kernel source and include files but nothing !!!!!
in /usr/include/bits/mman.h
there are some flags:
----
/* These are Linux-specific. */
#ifdef __USE_MISC
# define MAP_GROWSDOWN 0x0100 /* Stack-like segment. */ # define MAP_DENYWRITE 0x0800 /* ETXTBSY */ # define MAP_EXECUTABLE 0x1000 /* Mark it as an executable. */ # define MAP_LOCKED 0x2000 /* Lock the mapping. */ # define MAP_NORESERVE 0x4000 /* Don't check for reservations. */ #endif ----
no MAP_UNLOCKED present ...
is this a 2.3.x feature ?
Quoted 9 lines
Quoted 3 lines
What I'm trying to say:
read()doesn't work better thanmmap()in terms of behaviour during swapping, because the kernel could easily swap out your buffer where youread()in the data.Actually, the last time I tested this extensively (with a 2.0.x kernel I think),
mmap()was much better performing than read()..for a while. When sequential accessing of a file got to a point around my RAM limits, the disk activity became insane. Maybe things are better now.
Quoted 6 lines
Quoted 4 lines
I will add a feature to the pagefaulter thread which uses mlock()/munlock if root privileges are available, so that even
Eric is satisfied.
:-)Eric is hard to satisfy. :)
You are not the only !
:-)
Quoted 10 lines
Quoted 5 lines
PS:
mlockall(MCL_CURRENT|MCL_FUTURE)is a bad idea because when you do themmap()of the large file the process tries to load all into mem,and my scheme would not work. Better to usemlockall(MCL_CURRENT)and mlock()/munlock() areas on demand.This is a useful solution in some circumstances. However, it means I can't use malloc()/new, may have trouble with shared libraries or dynamically loaded plugins and have to worry about reserving stack space. It's not the way I want to work.
agreed for shared libs, but if you don't make any strange syscalls , I think there are not very much problems.
As for malloc : just mlock() the area after the allocation.
regards,
Benno.
Benno Senoner discourseth:
Quoted 5 lines
I think the best fallback solution is the following (but not 100% reliable in the case of syscalls):
mlock(MCL_CURRENT);mlock()malloec()ed aread
Hmm, I remain sceptical that I'll see large apps using this. Using other-people's-libraries is common after all. It would mean a lot of work that doesn't seem worth the benefit to me. However, it would be pleasant if people proved me wrong. :)
Quoted 2 lines
but strange,
MAP_UNLOCKEDseems to not exist on my 2.2.12 kernel !!
It doesn't exist..I made it up! :)
I think it would be A Good Idea, and it looks easy to add.
Quoted 4 lines
Quoted 1 line
Eric is hard to satisfy. :)
You are not the only !
:-)
Thence comes progress.
Eric