There was something like that in Kerrighed, a Linux kernel fork targeting clusters used among other things in XtreemOS in the 00s. See CAP_DISTANT_FORK here for instance.
It looks from that list as if they also have support for process migration (which means, unlike the approach in the article, they must be shipping kernel state for file descriptors elsewhere). If you have that, then telefork is trivial: like the original fork, you get the copy behaviour for free because you’re making a copy to do the migrate, you just need to not delete the original.
I’m not sure if it made it into the tree, but there was some more recent Linux work to allow hibernating processes: the entire process state, including everything necessary to reconstruct the kernel’s state associated with the process, was written to disk. Process migration was built on top of this (save process to disk, copy to remote host, rehydrate) and so telefork could be as well.
There was something like that in Kerrighed, a Linux kernel fork targeting clusters used among other things in XtreemOS in the 00s. See CAP_DISTANT_FORK here for instance.
It looks from that list as if they also have support for process migration (which means, unlike the approach in the article, they must be shipping kernel state for file descriptors elsewhere). If you have that, then telefork is trivial: like the original fork, you get the copy behaviour for free because you’re making a copy to do the migrate, you just need to not delete the original.
I’m not sure if it made it into the tree, but there was some more recent Linux work to allow hibernating processes: the entire process state, including everything necessary to reconstruct the kernel’s state associated with the process, was written to disk. Process migration was built on top of this (save process to disk, copy to remote host, rehydrate) and so telefork could be as well.