If I'm not mistaken, the John Dyson message referred to was this one, from comp.unix.bsd.freebsd.misc. It isn't quite as apocalyptic as the original AC post made it out to be...
Subject: Re: FreeBSD and SMP From: "John S. Dyson" Date: 2000/06/01 Newsgroups: comp.unix.bsd.freebsd.misc Bruce Burden wrote: > > XuYifeng wrote: > > : I am newbie of FreeBSD, I heard rumor that Linux 2.2 has better SMP support > : than FreeBSD 4.0? >: > I understand the latest Linux kernels have a multi-threaded > SMP kernel, while FreeBSD still uses the big lock model for SMP. > > From what little I have seen on the -current mailing list, > this is something the developers are going to have to "bite the > bullet" on, it seems, because the change isn't going to be easy > or compatible. >
About the time that I left FreeBSD, my biggest TECHNICAL argument was that FreeBSD needed an official project to re-design the kernel so as to support multi-processors correctly. In NO WAY does FreeBSD do multi-processing correctly (and that isn't an insult, but is a result of the monolithic kernel legacy -- FreeBSD is an excellent kernel otherwise!!!:-)). However, Linux supports SMP only marginally better than FreeBSD in the grand scheme of things.
A significant and well thought out redesign is in order. This is NOT an easy project and would require a serious committment on the part of the kernel developers. Hack solutions shouldn't even be considered for the shortest amount of time (the time involved in looking at expedient work-around type efforts is totally wasted.)
IMO, the Linux approach so far appears to be an iterative improvement on the same basic design: TRADITIONAL monolithic kernel. Given a traditional monolithic design, once the kernel is fine grained enough, and provides correct internal structure, the kernel maintenance people will have to be MUCH more intelligent than the original developers themselves. The complications in an adequate monolithic design are enormous.
The reason for my work being aborted on the FreeBSD SMP MONOLITHIC kernel wasn't that I couldn't make a FreeBSD SMP monolithic kernel work very, very well. The resulting kernel design would have been unmaintainable. I decided (when it was my place to decide a LONG time ago) that a monolthic SMP kernel (other than the current stopgap) didn't provide the quality that the FreeBSD name demands.
I am currently making really good and effective progress on my video compression (data reduction) software, and I might be able to do kernel hacking (actually redesign) in the near future. I have no need for employment anymore to make money, so my time is fully my own.
The last few years has been frustrating, but at least I worked with some tremendously nice people. I now have a situation that is very close to what it used to be when I first started working on FreeBSD!!! (I doubt that I'll be directly involved in the project because of disagreements that I had about the technical direction of the kernel, and marketing disagreements (with the associated burnt bridges:-(()).
I am going to be lurking much more than I have in the last 2yrs anyway:-). I just looked that the FreeBSD mailing lists a couple of weeks ago for the first time in a few years!!!
-- John | Never try to teach a pig to sing, dyson@iquest.net | it makes one look stupid | and it irritates the pig.
Re:Unstable Implementation (Score:5)
Subject: Re: FreeBSD and SMP :
:-)). However, Linux supports SMP only marginally better than FreeBSD in the grand scheme of things.
From: "John S. Dyson"
Date: 2000/06/01
Newsgroups: comp.unix.bsd.freebsd.misc
Bruce Burden wrote:
>
> XuYifeng wrote:
>
> : I am newbie of FreeBSD, I heard rumor that Linux 2.2 has better SMP support
> : than FreeBSD 4.0?
>
> I understand the latest Linux kernels have a multi-threaded
> SMP kernel, while FreeBSD still uses the big lock model for SMP.
>
> From what little I have seen on the
-current mailing list,
> this is something the developers are going to have to "bite the
> bullet" on, it seems, because the change isn't going to be easy
> or compatible.
>
About the time that I left FreeBSD, my biggest TECHNICAL argument was that FreeBSD needed an official project to re-design the kernel so as to support multi-processors correctly. In NO WAY does FreeBSD do multi-processing correctly (and that isn't an insult, but is a result of the monolithic kernel legacy -- FreeBSD is an excellent kernel otherwise!!!
A significant and well thought out redesign is in order. This is NOT an easy project and would require a serious committment on the part of the kernel developers. Hack solutions shouldn't even be considered for the shortest amount of time (the time involved in looking at expedient work-around type efforts is totally wasted.)
IMO, the Linux approach so far appears to be an iterative improvement on the same basic design: TRADITIONAL monolithic kernel. Given a traditional monolithic design, once the kernel is fine grained enough, and provides correct internal structure, the kernel maintenance people will have to be MUCH more intelligent than the original developers themselves. The complications in an adequate monolithic design are enormous.
The reason for my work being aborted on the FreeBSD SMP MONOLITHIC kernel wasn't that I couldn't make a FreeBSD SMP monolithic kernel work very, very well. The resulting kernel design would have been unmaintainable. I decided (when it was my place to decide a LONG time ago) that a monolthic SMP kernel (other than the current stopgap) didn't provide the quality that the FreeBSD name demands.
I am currently making really good and effective progress on my video compression (data reduction) software, and I might be able to do kernel hacking (actually redesign) in the near future. I have no need for employment anymore to make money, so my time is fully my own.
The last few years has been frustrating, but at least I worked with some tremendously nice people. I now have a situation that is very close to what it used to be when I first started working on FreeBSD!!! (I doubt that I'll be directly involved in the project because of disagreements that I had about the technical direction of the kernel, and marketing disagreements (with the associated burnt bridges :-(()).
I am going to be lurking much more than I have in the last 2yrs anyway :-). I just looked that the FreeBSD mailing lists a couple of weeks ago for the first time in a few years!!!
--
John | Never try to teach a pig to sing,
dyson@iquest.net | it makes one look stupid
| and it irritates the pig.