RepoMirrors/ceph
1
0
Fork 0
mirror of https://github.com/ceph/ceph synced 2024-12-26 05:25:09 +00:00
Code Issues Packages Projects Releases Wiki Activity

buffer: use raw_combined for certain allocations

Browse Source 
If the alignment is on a page boundary, or the allocation is big,
a separate buffer::raw goes faster.  The rest of the time,
a raw_combined does.

Signed-off-by: Sage Weil <sage@redhat.com>
This commit is contained in:
Sage Weil 2016-02-11 11:48:44 -05:00
parent 6be3b99d5d
commit 73dcd26fc8
1 changed files with 16 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
src/common/buffer.cc
View File

@ -709,13 +709,26 @@ static simple_spinlock_t buffer_debug_lock = SIMPLE_SPINLOCK_INITIALIZER;
}
buffer::raw* buffer::create_aligned(unsigned len, unsigned align) {
// If alignment is a page multiple, use a separate buffer::raw to
// avoid fragmenting the heap.
//
// Somewhat unexpectedly, I see consistently better performance
// from raw_combined than from raw even when the allocation size is
// a page multiple (but alignment is not).
//
// I also see better performance from a separate buffer::raw once the
// size passes 8KB.
if ((align & ~CEPH_PAGE_MASK) == 0 ||
len >= CEPH_PAGE_SIZE * 2) {
#ifndef __CYGWIN__
//return new raw_mmap_pages(len);
return new raw_posix_aligned(len, align);
return new raw_posix_aligned(len, align);
#else
return new raw_hack_aligned(len, align);
return new raw_hack_aligned(len, align);
#endif
}
return create_combined(len, align);
}
buffer::raw* buffer::create_page_aligned(unsigned len) {
return create_aligned(len, CEPH_PAGE_SIZE);
}