On 2014, Apr 11, at 12:30 PM, C. Bergström <cbergst...@pathscale.com> wrote:
> On 04/11/14 10:52 PM, Mark Hahn wrote: >>> I'm exploring tiny OS, in-house homebrew and anything non-heavy (aka linux) >>> for Xeon PHI. Does anyone on the list know of any open source work for >> >> I'm curious what you think the win will be. >> > Good question - I'd call this project more "for fun" than anything else at > this point. > > Hopefully it will allow us to easily test different on-card schedulers. (We > are already working on userland (host) scheduling). I've seen numerous cases > where "linux" doesn't scale well on many-core systems. > _______________________________________________ > Beowulf mailing list, Beowulf@beowulf.org sponsored by Penguin Computing > To change your subscription (digest mode or unsubscribe) visit > http://www.beowulf.org/mailman/listinfo/beowulf The MIT PDOS group has been fixing Linux, one scalability bug at a time, for years. Recently they have a new scheme for wholesale improvements, that is worth a look. See http://pdos.csail.mit.edu/commuter/ Nutshell - properties of the system call interface can make scalable implementations either possible or not. The data also suggests system calls to avoid. I'll argue that there is no real reason to think a new kernel will be more scalable than linux. It is relatively easy to make one smaller, but if you can fix the scalability problems, why not fix them for everyone? -Larry _______________________________________________ Beowulf mailing list, Beowulf@beowulf.org sponsored by Penguin Computing To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
