On 8/24/07, Jim Meyering <[EMAIL PROTECTED]> wrote: > Hi James, > > When you measured that, did the system you used have sufficient > openat support? If not, you probably saw the performance penalty of > the openat/fstatat/etc. wrapper functions that simulate the syscalls -- > probably via /proc/self/fd/%d/%s. If you run timings on a losing > kernel/libc combination, you should expect a performance penalty.
An additional effect was I think that some items in the directory tree were stat()ed a a number of times. I don't know how much of the performance difference that accounted for. James.
