>Maybe that's not such a good question (since I don't rightly >know what a true mu-kernel is -- but I'm curious...) A microkernel, by Tanenbaum's definition (page 388) only does 4 things: Interprocess communication (BMessages/ports, for us), some memory management, limited amount of low-level process management and scheduling and low level i/o. Without the source, it would be very hard to tell if Be's kernel was truly micro or not. And, of course, there is always a debate between whether add-ons make the kernel non-micro or not, and what "some" and "limited" mean. >How about: do all the servers run only as user-mode processes? Yes. >If so, would someone mind summing up what the whole >mechanism/policy concept means with regard to the >OBOS/BeOS client-server-kernel architecture? AFAIK, simply this. Stay out of kernel land. Little needs to be there. Mostly file systems and drivers. Servers do the "heavy lifting" - depending on who wrote what (style thing) the client can be anything from almost empty to fairly fat. Dianne indicated to me that BView's are > 5000 lines of C++ code on the client side. That seems like too much, to me. >(After reading Daniel's newsletter article and Tanenbaum's Modern >OS's chp.1, my interest is piqued. :) Tanenbaum is dense, but very good. I had it within arms reach. :-)