Re: [GSoC'13] Towards a Java-language unikernel substrate for Xen

[Anil Madhavapeddy <anil@xxxxxxxxxx>]

Thanks for investigating this, Shuang!  I'd be happy to supervise you once the 
GSoC applications open.

The general next goal should be to survey some of the other available 
unikernels (notably HalVM and GuestVM, as ErlangOnXen is closed source), and 
determine the set of libraries and device driver interfaces they expose. Mirage 
has a clean reimplementation of almost everything from the device driver up, 
and so should have the most minimal set of dependencies.  HalVM needs a limited 
libc, but also has a clean-slate TCP stack.

The Mirari wrapper should let us set a benchmark framework to run and evaluate 
the performance of the different stacks against stock benchmarks such as 
HTTPbench, on both UNIX userspace and Xen.  That'll point out hotspots and 
optimisation possibilities when it comes to building common libraries for all 
of them.

Meanwhile, do learn OCaml and feel free to ask questions on this list :-)

best,
Anil

On 6 Apr 2013, at 06:34, QIU Shuang <qiush.summer@xxxxxxxxx> wrote:

> Hi folks,
> 
> This is Shuang Qiu, a second year master student from Shanghai, China.
> 
> I'm warming-up for the coming Google Summer of Code 2013 and quite interested 
> in the proposed project in Xen community by Dr. Madhavapeddy "Towards a 
> multi-language unikernel substrate for Xen".
> Especially, I'm interested in the Java language part based on my previous 
> project experience and knowledge. (it's a pity I don't have a good knowledge 
> of OCaml currently, but I'm grasping it :p.)
> 
> I tested guestvm (https://kenai.com/projects/guestvm, the Guest VM 
> Microkernel sub-project), a Java unikernel substrate for Xen, on a Xen build 
> (xen-4.3-unstable). It works as expected. The simple hello-world case is just 
> print the system time every second.
> Just report the positive result and will read the guestvm code doing some 
> hacking later.
> 
> I also skipped the Mirage mail list archive and with the 1000-mile view of 
> project Mirage in mind. The project is promising. If a unikernel development 
> environment for multiple languages can be successfully setup, developing for 
> unikernel OS will be much easier and the efficiency of the dedicated running 
> process will be promoted in unikernel OS by bypassing the traditional 
> software stack in general purposed monolithic/micro- kernel OS. I'm seeking a 
> chance to make a little contributions to the community.
> 
> Thank you very much for your comments and cheers!
> Shuang