Peter Avalos has made it possible for TCP buffers to be automatically sized according to need, which improves performance when latency goes up. It’s already in.
The system leaf.dragonflybsd.org, where developer accounts are located, has had a significant bump in CPU and RAM, and has the newest scripts for automated vkernel setup.
As always, leaf accounts are available for anyone who wants to develop something on DragonFly, independent of commit access.
Release 2.3.2 has been tagged, for anyone who wants to stick with DragonFly past the current release but before the recent radical changes to NFS. Check the commit message for a summary of what’s changed since 2.3.1, taken from the commit messages.
Well, default as of the next release. With the recent changes to NFS, TCP will be needed in most circumstances. It’s still possible to force UDP if desired with the ‘mntudp’ option, or ‘udp’ if you’re using a 2.3.x system.
It is now possible to mount a Hammer PFS via NFS, though you’ll want to use NFSv3.
Peter Avalos has updated gzip, with a far longer list of updates than I would have thought possible with a utility that’s been around for a while.
Matthew Dillon made some major changes to NFS, which have greatly improved speed. He’s also made the clients able to write asychronously, which can overwhelm a server because of this increased throughput. Be careful.
Hey, look at what Michael Neumann’s doing: making Hammer expandable! It will be possible to expand your Hammer volumes while online, even.
(note: it’s experimental; don’t be surprised if it destroys data.)
DragonFly has its first 10G network driver, mxge(4), for the Myricom Myri10GE. Aggelos Economopoulos ported it from FreeBSD. Check his post for notes and credits for the people who helped out.
Michael Neumann has removed the PRISON_ROOT flag, and has changed jail(8) code to use only prison_priv_check() to check for allowed operations. This won’t mean anything from a user standpoint, but it does make programming easier.
pipe(2) is now MPSAFE, meaning it can take advantage of multiple processors without the Giant Lock. Matthew Dillon published some before-and-after stats in his commit.
Alexander Polakov has ported the ae(4) network driver from FreeBSD to DragonFly; it’s committed now. This device is common in some (many?) Asus Eee devices.
The kernel option PCI_MAP_FIXUP has been removed as of July 11th; if you’re upgrading past that point, make sure to remove that option.
avalon.dragonflybsd.org has a fresh set of pkgsrc-current binary packages for 2.2.1 located at http://avalon.dragonflybsd.org/packages/DragonFly-2.2/pkgsrc-current/. I’ll start a pkgsrc-2009Q2 build momentarily – the pkgsrc-2009Q2 build will become ‘stable’.
The short summary: everyone passed. Yay!
5 weeks to finish!
The in-progress code for the Summer of Code project ‘DragonFly on AMD64’ has been imported; you can now build for SMP on AMD64, and complete a installworld/buildworld, natively. Modules don’t (yet) compile…
Virtual kernels are now SMP by default on DragonFly, even if you don’t have multiple processors/cores.
Peter Avalos has updated libpcap to version 1.0.0 and tcpdump to 4.0.0. (tcpdump site) I’d guarantee that having at least a passing familiarity with tcpdump will eventually, someday, solve an otherwise intractable problem for you.
Threading libraries libc_r and libthread_xu have been synchronized by Hasso Tepper; this shouldn’t cause noticeable issues. The potential issues he mentions for pkgsrc appear fixed, as I haven’t had any significant trouble (from that, at least) during bulk builds.
Alex Hornung is looking for suggestions on the userland tool(s) for his devfs project. This is a Google Summer of Code project, and I’m a bit late posting this, so hurry if you want to get your two cents in.