UniLinux -- Unikernelized Linux Exploration
From: Tiejun Chen <[email protected]>
[ Note: The following abstract was submitted via the Linsux Plumbers
Conference website. Per the instructions that were posted for the Maintainer's / Kernel Summit Call for Proposals[1], the proposal should also be posted on the ksummit-discuss list, so that people can comment on the proposal, and perhaps start a discussion before the summit.
[1] https://lwn.net/Articles/788378/
Please note that topic proposals for both the Kernel Summit and the Maintainer's Summit are still welcome, and the deadline has been extended to June 3rd. -- Ted ]
Unikernel is relatively a novel software technology that links an application with OS in the form of a library and packages them into a specialized image that facilitates direct deployment on a hypervisor. And the reality is that no matter what we are talking the cloud infrastructure, or IoT edge systems, etc, we really need this across a variety of areas, buy those existing unikernels have yet to gain large popularity. So what if we can convert Linux to Unikernel or Unikernel-like? Here I'll talk what challenges Unikernels are facing, and discuss my proposal of if-how we could convert Linux as Unikernel like to make all kernel and user stuff working only kernel mode, how to manage interrupt, how to further reduce memory size and footprint, how to improve IO performance with light weight network stack, should we support multiple processes in some light weight ways, etc. These problem could be addressed but also need to further discuss during LPC.
In terms of UniLinux, IoT and Serverless could be a valuable candidate to make Unikernel succeed. Meantime, it should be a potential vessel by replacing container, because compared to traditional container, Unikernel is small, fast and more secure as a lightweight VM. So it is really worthy fitting Unilinux into a variety of areas like IoT, Serverless and so on.
Quoted 6 lines · Theodore Ts'o
From: Tiejun Chen <[email protected]>
Unikernel is relatively a novel software technology that links an application with OS in the form of a library and packages them into a specialized image that facilitates direct deployment on a hypervisor....
Tiejun, could you give some details regarding how your proposed presentation would differ from the one you gave at the OSS NA in Los Angelos?
https://ossna2017.sched.com/event/BDpy/unikernels-and-explorations-tiejun-chen-vmware
Many thanks!
- Ted
Hi Ted,
Looks I missed this email so really sorry for this delay response. When I went through ksummit-discuss today and just noticed this ☹
Overall, I didn't refactor too much into the big architecture of my proposal. But I'd like to 1>. go into something in detail with the participants ; 2>. talk a little bit new implementations. So besides of my original version, this time we would review-discuss them
1. Reduce the source code
For example, I'm thinking if we can remove unnecessary kernel parameter sections while compiling. Furthermore, we need to extend this to syscall.
2. Decouple the scheduler
Do we still need to keep all schedulers work in terms of those unikernel cases running one process? Even this process is running with one thread or few threads sometime.
3. Any refactors to some valuable use cases like a safety-critical application. What about classifying syscall to different levels?
4. Others
Instead of making out my solution simply during those summits, plus, each part in the original version is not be implemented generally, so I think ksummit is supposed to a good chance to help me address these challenges properly. So this time, in terms of my proposal, I'm open to any technical solutions.
Thanks
Tiejun
Quoted 31 lines
-----Original Message-----
From: Theodore Ts'o <[email protected]>
Sent: Thursday, May 30, 2019 9:55 PM
To: [email protected]
Cc: Tiejun Chen <[email protected]>
Subject: Re: [Ksummit-discuss] [TECH TOPIC] UniLinux -- Unikernelized Linux ExplorationQuoted 6 lines · Theodore Ts'o
From: Tiejun Chen <[email protected]>
Unikernel is relatively a novel software technology that links an application with OS in the form of a library and packages them into a specialized image that facilitates direct deployment on a hypervisor....
Tiejun, could you give some details regarding how your proposed presentation would differ from the one you gave at the OSS NA in Los Angelos?
https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fossn
a2017.sched.com%2Fevent%2FBDpy%2Funikernels-and-explorations-tiejun- chen- vmware&data=02%7C01%7Ctiejunc%40vmware.com%7C1e77cc17ac614 0465ab408d6e5065bed%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C0%7C0 %7C636948212810513960&sdata=hzZ9MpsXHx%2BEiAe%2BrJwXBjWiwx gAWdd8h26R6FcCwTc%3D&reserved=0Many thanks!
- Ted
Quoted 11 lines · Tiejun Chen
Hi Ted,
Looks I missed this email so really sorry for this delay response. When I went through ksummit-discuss today and just noticed this ☹
Overall, I didn't refactor too much into the big architecture of my proposal. But I'd like to 1>. go into something in detail with the participants ; 2>. talk a little bit new implementations. So besides of my original version, this time we would review-discuss them
1. Reduce the source code
For example, I'm thinking if we can remove unnecessary kernel parameter sections while compiling. Furthermore, we need to extend this to syscall.
2. Decouple the scheduler
Do we still need to keep all schedulers work in terms of those unikernel cases running one process? Even this process is running with one thread or few threads sometime.
3. Any refactors to some valuable use cases like a safety-critical application. What about classifying syscall to different levels?
4. Others
Tiejun, this looks like it's only vaguely about Linux. You may have started with the Linux kernel code at one point, but you've made sufficient changes that it's unclear to me how we could expect people who have never heard of UniKernel before to be able to participate in your these discussions.
If you had specific changes that you would like to be seen made to the mainline Linux kernel, perhaps it might be more profitable for you to make specific concrete suggestions, preferably backed up by code, and send it to the Linux Kernel mailing list, cc'ed to ksummit-discuss?
Regards,
- Ted
Quoted 37 lines
Instead of making out my solution simply during those summits, plus, each part in the original version is not be implemented generally, so I think ksummit is supposed to a good chance to help me address these challenges properly. So this time, in terms of my proposal, I'm open to any technical solutions.
Thanks
TiejunQuoted 31 lines
-----Original Message-----
From: Theodore Ts'o <[email protected]>
Sent: Thursday, May 30, 2019 9:55 PM
To: [email protected]
Cc: Tiejun Chen <[email protected]>
Subject: Re: [Ksummit-discuss] [TECH TOPIC] UniLinux -- Unikernelized Linux ExplorationQuoted 6 lines · Theodore Ts'o
From: Tiejun Chen <[email protected]>
Unikernel is relatively a novel software technology that links an application with OS in the form of a library and packages them into a specialized image that facilitates direct deployment on a hypervisor....
Tiejun, could you give some details regarding how your proposed presentation would differ from the one you gave at the OSS NA in Los Angelos?
https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fossn
a2017.sched.com%2Fevent%2FBDpy%2Funikernels-and-explorations-tiejun- chen- vmware&data=02%7C01%7Ctiejunc%40vmware.com%7C1e77cc17ac614 0465ab408d6e5065bed%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C0%7C0 %7C636948212810513960&sdata=hzZ9MpsXHx%2BEiAe%2BrJwXBjWiwx gAWdd8h26R6FcCwTc%3D&reserved=0Many thanks!
- Ted