53 lines (42 loc) · 1.5 KB

Architecture (prototype)

Goal

Build a minimal microkernel-ish prototype on i386 that demonstrates:

message passing (IPC)
services split out of the “core”
fault isolation benefits (service crash + restart)
performance comparison (IPC vs direct call)

Execution model

In week-1 scope, “services” run as separately scheduled tasks (kernel threads) with message queues.

If time permits, we can evolve to:

ring3 user mode + syscalls
user-mode servers
stronger isolation via page tables

Components

Kernel core

Boot + early init
Scheduler (cooperative OK)
Task abstraction (stack + state)
Basic drivers: VGA/serial output

IPC layer

Endpoints identified by small integer IDs
Messages are fixed-size structs or (type, len, payload)
Each endpoint has a bounded queue

Semantics (suggested)

ipc_send(dst, msg): enqueue or yield/retry when full
ipc_recv(src_filter, out_msg): dequeue or yield when empty
Optional: request/reply token for simple RPC

Service registry

Table: name -> endpoint_id
Boot-time registration for built-in services

Services

At least 3 services:

console/log service
echo service
timer service
monitor service (fault isolation)

Fault isolation demo

A service “crash” should not take down the whole kernel
A monitor restarts the service task and the system continues

Performance comparison

direct call loop vs IPC ping/pong loop
measure time using rdtsc if available, otherwise a rough tick-based approach