Gabriel and team,
On 24/10/16 16:00, Gabriel Parmer wrote:
We did all of this to demonstrate a scheduling technology that is likely of
less interest here. The performance of the system is roughly on-par with
Xen for nginx serving persistent connections even with a pretty brain-dead
data-movement mechanism that requires quite a bit of memcpy.
In the spirit of "one must learn everything to know just a little bit",
I'm personally interested in at least a short summary on your scheduling
technology. Furthermore, since the claim is that rump kernels are
scheduler-agnostic (which is really what separates them for everything
else), your work offers anecdotal proof, so I think at least some
further details are of general interest to this list. Links to papers
on your scheduler are of course also welcome, if you can share any
currently.
If there is any interest in integrating
this stuff into mainline, let us know and we'll see what's possible.
The interest depends on if your work can be seen to be useful outside of
Composite, what sort of maintenance burden it adds, etc. So, please
post details. Per the very least, like I [think I] said earlier, if you
want to propose hooks to "document" where your code attaches, those can
usually be added with relatively little pain.
Also, please make use of the wiki to link back to your work. There's
the obvious "publications" page:
http://wiki.rumpkernel.org/Info%3A-Publications-and-Talks
There's also the "who uses rump kernels" page, which I created a while
ago but forgot to advertise:
http://wiki.rumpkernel.org/Info%3A-Who-Uses-Rump-Kernels
Huge kudos to Robert Gifford, Luke Baier, and Phani Kishore Gadepalli for
pushing this work through.
Congrats! There's nothing quite like the feeling when something finally
works ;)
