index.md (2801B)
1 +++ 2 title = 'Kernels' 3 +++ 4 # Kernels 5 the type of kernel that you use, and the OS architecture, depends on the application 6 7 ## Monolithic kernels 8 9 - main program invokes syscall 10 - kernel is underlying monolithic block: 11 - service procedures carry out syscalls 12 - utility procedures help implement service procedures 13 14  15 16 - separate applications and OS using privilege levels into user and kernel 17 - on x86, 4 privilege levels (but in practice mostly 2 are used) 18 - this is supported by the hardware directly 19 - if only goal is to separate untrustworthy apps from lower level shit, you only need 2 separate levels 20 - if you include more levels, there’s a cost associated with switching between levels, so why do it if it’s not needed 21 22  23 24 ## Virtualisation 25 26 - originally to separate multiprogramming from extended machine 27 - N independent system call interfaces 28 29  30 31 - Virtual machine monitor (VMM/Hypervisor) emulates hardware 32 - types: 33 - 1: VMM runs on bare metal (like Xen) 34 35  36 37 - 2: VMM hosted on OS (like QEMU) 38 39  40 41 - Hybrid: VMM inside OS (like KVM) 42 43 ## Exokernel 44 45 - separate resource control from extended machine 46 - unlike VMM/Hypervisor, it: 47 - does not emulate hardware. only resource manager 48 - only provides *safe* low-level resource sharing 49 - service procedures are offered as library linked directly to application -- "Library OS" 50 - different library OSes for different programs, allows application-level specialisation 51 52 ## Client/server model (microkernel) 53 54 - organise service procedures in programs running in separate processes (system services/drivers) 55 - high level of isolation 56 - processes communicate via message passing 57 - calls rely on the same mechanism (message passing) 58 - messaging is implemented in microkernel (minimal kernel) 59 - principle of least privilege -- isolate every service in its own domain (address space, process, etc.) 60 - this is more secure, but lower performance (always a tradeoff). have to switch between modes and shit 61 62  63 64 ## Microvisor 65 66 - combination of hypervisor and microkernel 67 - different OS architectures have different design points, people look at convergence and tradeoffs 68 69  70 71 ## Unikernel 72 73 - "single simple application implementing whatever in the cloud, in most efficient way" 74 - squash application and OS kernel into one thing, don't need all of the other stuff like process management 75 - gets rid of all of the overhead