FreeBSD and Linux

Fare Rideau
Thu, 1 May 1997 12:32:29 +0200 (MET DST)

> Well, the FreeBSD kernel is also an option.  I was thinking about
> the pros and cons of each OS/kernel.

> [Licenses, Documentation (particularly networking), kernel threads]

* Licenses is a non-issue, as I've explained in my previous message.
* Linux networking seems pretty straightforward,
 much like anything in Linux, BTW; Linux is not so badly documented
 (a book by Remy Card will be out RSN, if it ain't already),
 and in the last case, the people in the various Linux kernel mailing lists
 will be happy to explain you how it works.
 So again, I don't think Documentation is an issue in our choosing
 between BSD and Linux.
* Kernel threads: fork() and mmap() already allow to share memory.
 There remains sharing of file descriptors.
 Well, seeing that file operations are a bottleneck,
 requiring that only one process may do I/O, and all others go through it
 seem ok; implementing this kind of thing in software will be needed
 somewhere along a Lisp kernel project anyway.
 And with Lisp, automatically generating the proxies from the list
 of primitives looks trivial...

All in all, I see no reason here for a choice between FreeBSD and Linux.
Not that there might not be a good reason in the future.
Meanwhile, let's go ahead and code: if there's a reason to choose,
that's the best way to find out!

> But I think for scalability, we definitely need the kernel
> threading.
I'm not convinced. I don't think that blazing fast SMP performance
be an immediate goal for the LispEnv project.
The difficulty of an efficient SMP GC is enough
to make me think we would better tackle the problem
when we have an all-LISP environment,
rather than by using the lame C we'd have to use if doing that now.