Hello, from my limited understanding of kernels I would think its like this.
The fastest operating system on earth would be one which never checks any parameters and would only have one user which can do anything (this saves doing privlege checks). (Hmm isnt this a description for windows95, except that it never was fast?) But the problem whith this operating system is that it will crash whenever you pass wrong parameters. (Ahh yes thats definitely windows) Next comes something like Gnu/Linux which has different privilege levels and checks parameters for reasonable values. This has of course an impact in the performace. And then there is something like Gnu/Mach which splits the work into different usermode servers which can be restarted independently if they crash. With a kernel like Mach you are actually slowed down because you always have to send IPC messages and do a context switch back and forth to the server from which you want information because you can't just read the information from the kernel memory like most of the linux systemcalls do. So yes it will be slower as long as IPC and the context switch is slower than reading directly from the kernel memory. Therefore getting IPC and context switches faster is one of the research topics in the so called second generation microkernels. And they got it really fast in kernels like L4. I think in one of the papers they describe how they run a minimal modified Linux kernel as a server on top of L4 and had only a performance penalty of about 6-9% If this could be improved in the future or how fast it will be on specialized hardware is an ongoing research I think. So its up to you to decide how much computing power you want to spent for the os and then you can choose between them. My personal opinion is that it is defineately worth because of security, user extendability and well defined interfaces! Please correct me if some of my assupmtions I have done here are wrong. Sincerly Fabian On Thu, Jul 25, 2002 at 01:29:20AM +0300, Catalin wrote: > Hello, > I have a question for you: > An low level application that runs under HURD should first "talk" with a > server and the server will send the request to the microkernel and the > mircokernel to the hardware. In linux the same application "talks" > directly with the kernel and the kernel with the hardware. But in hurd > if an error occurs inside the server then that particulary server > colapses but the mircokernel and the other servers survive(and can even > restart that server), and if the same kind of error ocurs in the linux > kernel(or other bigger kernel) the kernel colapses. > So the HURD will be the most stable and the most secure operating system > on earth but also the most slowest operating system on earth (because > the request travels a longer way)? > Please help me clear this image I have about the "microkernel versus > kernel".... > > Catalin -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
