Unless I am mistaken, they are re-inventing a Microkernel architecture, except that there is no central microkernel: it's a network of peers in a heterogeneous environment.
This is an operating system for supercomputers, something Microsoft has a very hard time competing in. Linux runs on 88% on all top 500 machines... and Windows on 1%: http://www.top500.org/stats/list/33/osfam http://www.top500.org/stats/list/33/os This is not an easy task. There is a reason why GNU Hurd (microkernel-based) has been in development since 1990... It's VERY hard to debug a concurrent system. You get all sorts of problem like Deadlocks and Race Conditions that are very hard to reproduce and even harder to figure out. I've also noted that theirs tests seems to indicate that their approach for memory allocation give faster results than Windows at around 7 cores and faster than a Linux-based system at around 13 cores. This is an academic curiosity... for now. A very interesting one :) David Montminy Alex Gal wrote: > This is interesting. From their introduction in sosp.pdf: > > "Moreover, these optimizations involve tradeoffs specfic to hard- > ware parameters such as the cache hierarchy, the memory consis- > tency model, and relative costs of local and remote cache access, > and so are not portable between different hardware types. Often, > they are not even applicable to future generations of the same archi- > tecture. Typically, because of these diffculties, a scalability prob- > lem must affect a substantial group of users before it will receive > developer attention. > We attribute these engineering diffculties to the basic struc- > ture of a shared-memory kernel with data structures protected by > locks, and in this paper we argue for rethinking the structure of > the OS as a distributed system of functional units communicat- > ing via explicit messages. We identify three design principles: (1) > make all inter-core communication explicit, (2) make OS structure > hardware-neutral, and (3) view state as replicated instead of shared. """ > > So shared memory is replaced by message passing. (reminds me a bit of > threads with shared data vs queues and multiprocesses) > > Any linux hackers care to discuss if this seems like a good idea ? > > > > On Sun, Oct 4, 2009 at 9:32 PM, Leslie Satenstein > <[email protected]> wrote: >> It is organized such that the open source kernel can use closed source >> (commercial) packages, without the constraints of the GNU (FOSS) license to >> make source available. >> >> The system will probably be open to wider experimentation in the world >> community sometime in 2012. >> >> >> On Sun, Oct 4, 2009 at 8:13 PM, Valery Shaevitch <[email protected]> wrote: >>> Hi, list and everyone here ! >>> >>> Microsoft Cambridge Research just has announced a first MS-supported >>> open source kernel developed at http://www.barrelfish.org/ >>> >>> I should say WOW ! >>> Seems like M$ is going right now >>> >>> Let us your opinion. >>> Who has successfully compiled the kernel ? >>> I do not have the Haskel compiler installed yet >>> Anyone ? >>> >>> _______________________________________________ >>> mlug mailing list >>> [email protected] >>> https://listes.koumbit.net/cgi-bin/mailman/listinfo/mlug-listserv.mlug.ca >> >> >> -- >> ------------------- >> Regards >> >> Leslie Satenstein >> >> _______________________________________________ >> mlug mailing list >> [email protected] >> https://listes.koumbit.net/cgi-bin/mailman/listinfo/mlug-listserv.mlug.ca >> >> > _______________________________________________ > mlug mailing list > [email protected] > https://listes.koumbit.net/cgi-bin/mailman/listinfo/mlug-listserv.mlug.ca _______________________________________________ mlug mailing list [email protected] https://listes.koumbit.net/cgi-bin/mailman/listinfo/mlug-listserv.mlug.ca
