Hi, I've been looking in to testing d_alloc_parallel() for correctness and performance after making changes to it. I haven't found anything useful. I hacked one .c file to fill a directory with other directories and then I hacked another .c file to walk/ scan that directory with multiple threads. I don't see much wakes (due to DCACHE_PAR_LOOKUP) on shm. This is different on xfs. The runtime of the walk/ scan directory varies a lot so I can't spot a difference between orig vs patched. On shm I have a span between 15 and 55 secs. Any suggestions? Sebastian
