FreeBSD 5.3-RELEASE on master FTP servers

[bascule]

Just thought I'd point out that the builds of FreeBSD 5.3-RELEASE have been uploaded to ftp://ftp.freebsd.org/pub/FreeBSD/releases/

We should see a release announcement today or tommorrow, hopefully, which will make the release official. They may swap files around before then, so downloader beware!

It's been a long haul for us FreeBSD folks from 4-STABLE to 5, but FreeBSD now runs beautifully on SMP systems, and can sport a constant time scheduler much like the one in Linux 2.6 (and Solaris) or kernel preemption, available in Linux, Windows NT, or Solaris. It now employs a highly scalable M:N threading model. Furthermore, the base system is substantially cleaner and more reliable, and none of the base system utilites depend on Perl any longer, so Perl is managed entirely through the ports collection.

[Gadsden]

I wonder how well this release will run on vmware? I think it was 5.1 that I tried last.. caused vmware 4.5 to blow up real good...

[skroo]

I wonder how well this release will run on vmware? I think it was 5.1 that I tried last.. caused vmware 4.5 to blow up real good...

I'm really interested in how the SMP is working out for them. From what bascule's said, it sounds like they went with a pretty interesting model.

[bascule]

I'm really interested in how the SMP is working out for them. From what bascule's said, it sounds like they went with a pretty interesting model.

They aren't quite as far along as Linux in getting rid of their Giant lock across the whole kernel. You can view the progress here:

http://www.freebsd.org/projects/busdma/

However one thing they did do was design an adaptive Giant mutex which, as I understand it, is actually implemented as a number of finer graned locks across various kernel subsystems, and the Giant lock is able to select which subsystems need to be locked based on what operations the driver is performing, so they can still get excellent SMP performance out of drivers which haven't been converted to the new busdma API.

Otherwise, the new M:N threads implementation provides excellent scalability across multiple processors, especially when used in conjunction with the new constant time ULE scheduler. All in all, it's pretty much on par with Linux 2.6 and the NPTL threads library.

[bascule]

Here's the official release announcement:

http://news-reader.org/article.php?g...ce&post_nr=441