I checked in some code yesterday to rlm_files that gets rid of 99.9% of the linked list walking. In my tests with 200k entries in a "users" file, the CPU time goes from large to negligible. The code changes are pretty trivial, which is a little surprising. It's based on the new growable hash functions, so most of the magic is hidden from the module. Is there any need for rlm_fastusers any more? It was always a bit of a hack, and with rlm_files being fast now, there doesn't appear to be much need for rlm_fastusers. I don't even know if anyone is using it... About the only difference is the reload support. I think the solution is to add a "need_reload" entry to each module, or maybe a "reload". The idea would be either that on HUP, the module would be able to say "no files change, I don't need a reload", or the module would be able to selectively reload the files that have changed. Comments? Alan DeKok.