On 12-05-30 01:11 PM, Raphaël Hertzog wrote:
>
> sync_file_range() is something different, it just tells the filesystem to
> start the writeback, it doesn't wait for it to have happened (contrary to
> fsync()).
>
> So it should be irrelevant for the time lost, unless Linux's NFS driver
> has a poor handling of this Linux-specific function.
Well, I tried NOOPing it with an LD_PRELOAD:
sync_file_range(int fd, off64_t offset, off64_t nbytes, unsigned int flags) {
printf("sync_file_range()\n");
return 0;
}
but that doesn't seem to have helped performance any.
On 12-05-30 01:11 PM, Raphaël Hertzog wrote:
>
> sync_file_range() is something different, it just tells the filesystem to
> start the writeback, it doesn't wait for it to have happened (contrary to
> fsync()).
>
> So it should be irrelevant for the time lost, unless Linux's NFS driver
> has a poor handling of this Linux-specific function.
Well, I tried NOOPing it with an LD_PRELOAD:
sync_file_range(int fd, off64_t offset, off64_t nbytes,
unsigned int flags) {
return 0;
}
but that doesn't seem to have helped performance any.