I'm curious why the concept of scheduling niceness applies only to an entire process, and it's not possible to have nice threads within a process. Is there any fundamental reason why it couldn't be supported with some extra bookkeeping to track niceness per thread? -- Ian