Aaron LI has brought dhcpcd into DragonFly-current, as a replacement to dhclient(8) and rtsold(8). dhclient does support IPv6 but it’s bundled with server and BIND applications, so it’s harder to support.
There’s a fix for memory contention in NUMA (meaning Threadripper in this case) configurations on DragonFly; the commit has before-and-after numbers. They are somewhat context-free, so I can’t easily translate to what this means for performance.
Matthew Dillon’s moved tty_token from a global to per-CPU token in most cases in DragonFly. This is good for performance as with any global->local shift, but I can’t tell you what aspect it improves.
If your kernel panics, the current state of memory can show why. That memory dump needs to be saved somewhere. ‘dumpon’ is the command to specify the device that will keep it. If you want to turn it off, you end up using the odd syntax ‘dumpon off’. Thanks to Aaron Li, there’s now a linguistically-sane command: ‘dumpoff’.
As a fix for wpa_supplicant, the upper limit on socket datagrams has been increased. What else does this affect? We’ll find out the hard way, which is why I mention it here.
Chromium, the open sourced base of the Chrome browser, builds on BSDs, including DragonFly. But not without some work.
DragonFly’s root account defaults to tcsh, and that now defaults to autorehash being set on. Useful to remember if you reflexively type ‘rehash’ like I do, and also useful if you come from a shell where ‘rehash’ isn’t needed.
DragonFly’s default compiler is now gcc-8. This will help with some amount of dports builds.
Matthew Dillon (re?)added a sysctl: vfs.hammer2.cluster_write. It defaults to off, since HAMMER2 already writes a large buffer size and this should, in theory, not be needed. It may improve performance in some situations where there’s a lot of file creation and deletion, but that’s my theoretical guess rather than anything I’ve bennchmarked.
If you haven’t done it before, you can use ‘make rescue’ to build a tiny base system on DragonFly, for use when /usr goes missing, for when your disk is encrypted, and other rather catastrophic problems. It should be in sync with the rest of the system, which is why ‘make rescue’ can be part of a buildworld process. I’m mentioning this because currently, ‘make upgrade’ should be done first.