D Kernel Project Proposal

Student/developer: Daniel Benamy

Professor/manager/mentor: Jason Nieh

Personal Motivation

My main motivations for this project are two-fold. First, the D language looks like an interesting idea and I've been interested in writing a sizeable system using it for a while. Second, the idea of writing my own kernel from scratch is very appealing to me. Whenever I work on a low level project, some bit of computers which used to seems a bit like magic, becomes clear to me. This gives me a better picture of how computer systems work and, I hope, makes me a better software engineer and computer scientist.

Why Interesting

There are some interesting questions which this project may be able to answer. One of the advantages of D over C and C++ is that it has some safety and productivity features which generally are seen in higher level languages such as (optional) garbage collection, bounds checked arrays, and built in support for automated testing of code. I imagine that using these will have an impact on the performance of the kernel, but they will probably also allow for faster and more bug free code. Examining this tradeoff will be interesting. Once the basic kernel works, I'll run some benchmarks to try to quantify how much slower it is than a similar kernel in C. This will probably be an imprecise comparison, but may be useful anyway. It would also be interesting to know how the development speed and quality compare to a kernel written in C although I don't have any idea how to measure that.

Development Goals

The schedule contains some detail about what I hope to accomplish, but the basic idea is simple: a minimal unix-like kernel with userspace processes. Since I'm hoping to do as much as possible myself from scratch to maximize my learning experience, I don't expect to get very far towards building a usable system. Also, since there are many unknowns, I may have to adjust the schedule as the project progresses.

I'm going to be targetting 32 bit x86 using QEMU. I'm going to assume the machine has some basic BIOS functions I can use for hardware access.

I don't have the development of a real filesystem on my schedule. Instead, I'm copying an idea out of CMU's kernel project: I'm going to have a couple of files available by encoding them as binary arrays and compiling them into the kernel. Then by reading certain memory locations, it'll have access to those files. If I manage to be ahead of schedule or don't need my buffer week for anything else, I'd love to try to implement a simple real filesystem or maybe port something over from a microkernel.

Name

There is an existing d kernel project is called "d kernel" so I need a new name. Until something better comes along, I'm calling it Dunk.