OSNews

You can sandbox user space apps without them, but the goal of microkernels is to also sandbox device drivers and the OS itself, so as to reduce the trusted computing base for a given task.

Considering how easily a buggy kernel-space driver can crash/lockup the whole OS or worse (I should know, I periodically experience that with GPU drivers and a USB piezo motor controller which we use at work), that's a Good Thing.

And this is why almost all OSs which initially went all monolithic are trying to push stuff back to user space nowadays. See WDDM drivers on Windows or FUSE on Linux as an example. I doubt that they will ever manage a full transition though, there's just too much work to do on a full-grown OS.

EDIT: As a bonus, microkernel also provide "natural" design guidelines for user app sandboxing. You can start by having security permissions that follow system component boundaries such as "Can use system components X, Y and Z", which are essentially free to implement. Then, you iterate from there towards finer-grained and coarser-grained mechanisms where needed. With monolithic architectures, on the other hand you are generally forced to derive and implement your sandboxing rules from the vacuum.

Edited 2013-02-07 06:24 UTC

Yes, most mobile OS are approaching micro-kernel architectures at least at user space level.

Mac OS X sandboxes with XPC communication is also quite similar to how micro-kernels work.

QNX and Symbian have showed that it is possible to achieve a high throughput when communicating between tasks, by moving pointer/handle ownership instead of copying data.

Another approach would be Microsoft's research to use the operating system as a library on top of an hypervisor, Drawbridge. From the same research group that created Singularity.

http://research.microsoft.com/en-us/projects/drawbridge/default.asp...