Microkernel question

[GWestphal]

I've been reading some of the Torvalds Tanenbaum debates and I saw someone say that microkernels can't share memory as each "server" has it's own address space.

Can't you just create a server explicitly for shared addressing? Have an application request access to said shared addressing server? Since all these servers address space is in RAM isn't it super low latency, wide bandwidth to have inter-server communication anyways?

 

[Ancalagon44]

I havent read the debates, but I have a funny feeling that server in this context refers to services and daemons, not an actual physical server. In other words, other processes running on the same machine as the kernel.

Care to post the debate? At least a link to the text? My thinking is that you dont normally want to share memory anyway, since it will just become corrupted and useless. Inter process communication is quite a science for exactly that reason, and the problem would become even worse over a network. Hence, nobody really wants two or more processes to share memory or address spaces, but they may want them to be able to pass information between them. That is a very different idea though.

 

[GWestphal]

Each server is a critical function, like filesystem or device drivers. Though with a microkernel technically you can have multiple simultaneous OS running.

http://slashdot.org/story/06/05/10/0439246/torvalds-on-the-microkernel-debate