DragonFly version 4.6 brings more updates to accelerated video for both i915 and radeon users, home-grown support for NVMe controllers, preliminary EFI support, improvements in SMP and networking performance under heavy load, and a full range of binary packages.
It is nice to see new releases of highly technical projects as DragonFly BSD, ReactOS or GenodeOS.
Kudos for them.
This is huge for me. I run a DragonFly VM under KVM with 32GB RAM and virtio as a test box to make sure I’m not accidentally introducing Linuxisms, so this is great news for me.
tidux,
I’m guilty of that. I’ve been using epoll/eventfd alot. The problem is that posix AIO is very poorly supported on linux because it’s implemented with blocking threads, which defeats the entire purpose of AIO. Since I’m not familiar with it, how is the state of AIO on BSDs? I am interested in hearing about what kind of things would typically break for you?
I’ve heard that kqueue is much nicer than epoll, but I haven’t written anything that needs either so I can’t comment on it directly. Using DragonFly BSD for testing uncovered some surprising issues with existing projects written in Go.
I found this comparison, which favors kqueue as well.
https://people.eecs.berkeley.edu/~sangjin/2012/12/21/epoll-vs-kqueue…
Apparently kqueue can handle timers, events, and disk IO. Epoll is limited to sockets, and they’ve invented new kinds of sockets like eventfd and timerfd so that those things could be polled as though they were sockets. Linux doesn’t support AIO for disks/files at all, which is why Posix had to handle it the way they did.
Another review with similar conclusion:
http://austingwalters.com/io-multiplexing/
So yea while I haven’t (and can’t) use it, kqueue might be the preferable API.
Edited 2016-08-04 23:20 UTC
I would love to see a comparison on the HAMMER fs and its default settings against others like zfs and privacy implications. Not from a crypto standpoint, but rather from a standpoint on general knowledge of what that filesystem is doing behind the scenes. When I played around with it in the past, I found default snapshot directories that even though snapshots were eventually turned off, the old files still existed, even after a hammer cleanup. The deleted directories could not be erased. This is not HAMMER 2 I’m talking about.
It seems to have some technical merits as a server platform.
I wonder if there is any compelling reason on the horizon for it to become an alternative on the desktop.