munmap() slowness
munmap on a 15TB anonymous area is extremely slow on our Itanium 2 box.
#include <sys/mman.h>
int main(void) {
size_t s = 1024*1024*1024*1024L * 15;
munmap(mmap(0, s, PROT_READ|PROT_WRITE,
MAP_NORESERVE|MAP_PRIVATE|MAP_ANONYMOUS, -1, 0),
s);
return 0;
}./a.out 0.00s user 1.51s system 100% cpu 1.515 total
The kernel is 2.4.19 plus linux-2.4.19-ia64-020821.diff.gz (plus some local changes that shouldn't be relevant).
Of course this isn't a real app, but I thought this was an interesting result.
Please CC me on any replies, as I am not subscribed to the list.
-andy
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Andy Isaacson
Andy> munmap on a 15TB anonymous area is extremely slow on our Itanium 2 box. Andy> #include <sys/mman.h>
Andy> int main(void) {
Andy> size_t s = 1024*1024*1024*1024L * 15;
Andy> munmap(mmap(0, s, PROT_READ|PROT_WRITE,
Andy> MAP_NORESERVE|MAP_PRIVATE|MAP_ANONYMOUS, -1, 0),
Andy> s);
Andy> return 0;
Andy> }
Andy> ./a.out 0.00s user 1.51s system 100% cpu 1.515 total
Andy> The kernel is 2.4.19 plus linux-2.4.19-ia64-020821.diff.gz (plus some Andy> local changes that shouldn't be relevant).
Andy> Of course this isn't a real app, but I thought this was an interesting Andy> result.
Andy> Please CC me on any replies, as I am not subscribed to the list.
$ uname -r 2.4.19-mckinley-smp $ /usr/bin/time ./a.out 0.00user 0.00system 0:00.00elapsed 100%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (44major+13minor)pagefaults 0swaps
strace output says:
mmap(NULL, 16492674416640, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, -1, 0) = 0x20000000002b4000 munmap(0x20000000002b4000, 16492674416640) = 0
Seems to perform just fine.
--david
Quoted 12 lines · David Mosberger
Andy> munmap on a 15TB anonymous area is extremely slow on our Itanium 2 box. Andy> #include <
sys/mman.h>
Andy> int main(void) {
Andy>size_ts = 1024*1024*1024*1024L * 15;
Andy> munmap(mmap(0, s, PROT_READ|PROT_WRITE,
Andy> MAP_NORESERVE|MAP_PRIVATE|MAP_ANONYMOUS, -1, 0),
Andy> s);
Andy> return 0;
Andy> }Andy> ./a.out 0.00s user 1.51s system 100% cpu 1.515 total
Quoted 8 lines
Andy> The kernel is 2.4.19 plus linux-2.4.19-ia64-020821.diff.gz (plus some Andy> local changes that shouldn't be relevant).
$ uname -r 2.4.19-mckinley-smp $ /usr/bin/time ./a.out 0.00user 0.00system 0:00.00elapsed 100%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (44major+13minor)pagefaults 0swaps
My bad, it turns out I logged into a different machine which is running 2.4.18-e.12smp from some interim RHAS 2.1 release. When I re-ran the test on the correct 2.4.19 box, the performance is fine.
-andy