Due to one of my classes in school, and mainly personal interest, I need to do research on a topic. I have an interest in operating systems, so I want to do research on operating system design. I have been trying to find and read existing research on this topic (experimental design for both for entire kernels or for some subsystem in the kernel). From this, I am hoping to be able to find an area that has not been researched in detail yet, and I want to do research on this (e.g. performance compared to standard designs). Also, I have a bit of experience with basic operating system development and have been working on my own hobby operating system for around a year now, and from this I have what I feel is a basic understanding of the main concepts in operating systems.
The reason I am making this post is to inquire whether there are any resources I can use to find these gaps in research easier / quicker, or places where I can go to get more advice on this topic, in case this type of post isn't for this site.
Operating systems research advice
-
Demindiro
- Member
- Posts: 165
- Joined: Fri Jun 11, 2021 6:02 am
- Libera.chat IRC: demindiro
- Location: Belgium
- Contact:
Re: Operating systems research advice
I believe some underexplored areas are:
Multikernels
The basic idea is: instead of one kernel for an entire system, have one kernel for each core of a system and use message passing between each kernel. It is aimed at addressing scalability problems as more and more cores and memory is crammed into a single system. I believe it would also be a very good fit for heterogenous processors (e.g. BIG.little) as with some processors each core may have different feature sets and/or accelerators.
Barrelfish OS seems to have coined the term, so that is a good starting point.
IOMMUs
IOMMUs have been around for a while, yet I feel no kernel takes full advantage of it. Micro- and exokernels in particular could benefit from it as it allows software to access hardware directly while still enforcing isolation between processes.
Hardware capabilities
Basically: tagging pointers with permissions and a length. CHERI is a notable example that is mostly compatible with software written for current architectures, but you can to more extreme lengths and e.g. enforce a distinction between addresses and "normal" values. It is closer to HW (or VM) than OS design though.
----------
If I had to pick one of these topics I'd go for multikernels. I think it'd be the most rewarding in terms of effort, though I wouldn't expect too much from any of the topics I listed.
- Multikernels; and
- Kernels designed around IOMMUs
- Architectures with hardware capabilities.
Multikernels
The basic idea is: instead of one kernel for an entire system, have one kernel for each core of a system and use message passing between each kernel. It is aimed at addressing scalability problems as more and more cores and memory is crammed into a single system. I believe it would also be a very good fit for heterogenous processors (e.g. BIG.little) as with some processors each core may have different feature sets and/or accelerators.
Barrelfish OS seems to have coined the term, so that is a good starting point.
IOMMUs
IOMMUs have been around for a while, yet I feel no kernel takes full advantage of it. Micro- and exokernels in particular could benefit from it as it allows software to access hardware directly while still enforcing isolation between processes.
Hardware capabilities
Basically: tagging pointers with permissions and a length. CHERI is a notable example that is mostly compatible with software written for current architectures, but you can to more extreme lengths and e.g. enforce a distinction between addresses and "normal" values. It is closer to HW (or VM) than OS design though.
----------
If I had to pick one of these topics I'd go for multikernels. I think it'd be the most rewarding in terms of effort, though I wouldn't expect too much from any of the topics I listed.
Re: Operating systems research advice
@Demindiro Thank you so much for the guidance man! Truly appreciate it 