FreeBSD 2020 Q4 Status Report | Hacker News Reader

FreeBSD 2020 Q4 Status Report

freebsd.org

156 pointsthrow0101a47 comments

Didn't anyone else notice this? NFS over TLS ? Coult it also do tls client certificate authentication?

EDIT: Apaprently yes, it can.

This is a game changer... I'm looking forward to have this in my TrueNAS box.

In the meantime... I might give FreeBSD another try.

It may be a while before it works on TrueNAS. NFS over TLS is a new feature for FreeBSD-13, as is kernel TLS We're still debating what to default ktls to (and have settled on "compiled in, but disabled via sysctl").

Here is the 'HOWTO' for it:

https://people.freebsd.org/~rmacklem/nfs-over-tls-setup.txt

FreeBSD is an amazing project. One of the changes I'm most excited about is they recently updated their /bin/sh (based on almquist shell) to support the Actually Portable Executable format used by the Cosmopolitan C Library. See https://github.com/jart/cosmopolitan/issues/23#issuecomment-... and the recent HN thread here: https://news.ycombinator.com/item?id=25556286

Is anyone aware of the status of support for 802.11ac?

I really do like some of the BSD’s but I always find myself having to purchase external wireless g/n dongles when I try to use it on recent machines. I recall a while back hearing that there was an effort to support it. I’d love to help too, but by the time I’d be able to familiarize myself with the FreeBSD kernel and various 802.11 standards it’d likely be finished lol.

At least for Intel support, look in the wireless update section of that page. You can even find out whom to contact if you want to help. :)

Oh whoops, somehow I completely overlooked that.

Anyone here use FreeBSD on the desktop? If so, what window manager and login manager do you use? Does Zoom work? I tried using Freebase a year or two ago and found it a bit painful, so I switched back to Fedora...

My desktop is a Threadripper 2990WX running FreeBSD-current with Nvidia graphics.

Its pretty usable as a desktop. I recently switched from KDE/Plasma5 to LXDE, mostly to avoid rolling-release related churn. Eg, I don't want my UI to change everytime I update firefox.

I tried and failed to use it for Google Meet quite a while ago. The problem I had was something to do with the AMD USB chipset. I need to try it again.

What doesn't work natively is anything with DRM. There are workarounds, like running Linux chrome in a linux jail (think WSL), or running spotifyd for Spotify.

How much trouble is getting Nvidia to work on a BSD? I'm considering switching but I have a similarly high-end set up and it would be a shame to let that hardware go to waste.

Its far less work than AMD or Intel drivers which depend on a massive Linux KPI shim layer.

The nvidia drivers are a binary blob surrounded by a FreeBSD src-based shim layer. If you're running a -release branch, you just install the nvida driver pkg and you're done. If you run -current like I do (with no KBI guarantees) you re-build the nvidia port every time you update your OS.

All in all, the nvidia drivers "just work" and are very boring. I'll pay a premium for boring.

NVIDIA provides drivers for FreeBSD, like they do for Linux.

The NVIDIA drivers work fine for graphics applications.

Nevertheless NVIDIA CUDA is available only for Linux, so NVIDIA GPU compute applications do not work directly, but maybe only using the Linux compatibility layer (I have not tried that).

Nvidia is completely boring and predictable.

Which is exactly what you want for a reliable compute box.

There used to be PC-BSD and DesktopBSD that were out-of-the-box desktop distributions of FreeBSD. I used them on my laptop through college and loved them - I feel like the end of those projects has left a bit of a hole in the community.

I am mistaken: 2 or 3 of the projects named here have recent updates and they look good. I might have to play with these this week: https://www.truenas.com/trueos-discontinuation/

I'm also using FreeBSD 12.2 Stable with KDE and SDDM. The initial setup is pretty straightforward, but I agree that every now and then I it can get really annoying. Today I tried to install VirtualBox 6 and realised that it installed v5, even though v6 is already more than a year old. Even the ports included v5.

I'm using FreeBSD on my laptop with i3 wm and no login manager. Zoom's sound doesn't work.

Some questions...

1) I use Zoom on Linux, and usually there's no sound until I go to the Zoom sound settings, press the "test sound" button, and then Output sound gets enabled. I have to do it for every meeting. Have you tried that?

2) Are you using Pulseaudio?

With xubuntu 18.04 and 20.04 I’ve always had sound right away in zoom. So there’s definitely some way to avoid that bug on linux.

Do you use startx to fire up i3?

I do. Pretty normal/simple, IIRC

I also use FreeBSD with i3. However, I dual-boot to Windows to use Zoom. Here's my write-up about videoconferencing options on FreeBSD: https://www.davidschlachter.com/misc/freebsd-videoconferenci... You can use Zoom in a browser and it works OK.

I normally use FreeBSD on servers, so I have little experience with desktops.

Nevertheless, a couple of years ago I have used FreeBSD also as a desktop for some time.

I have used XFCE, installed by compiling from sources, from FreeBSD ports, because that is what I use on Linux and I did not notice any problems.

The desktop experience depends a lot on what hardware you have, because of possible compatibility issues.

With an AMD Radeon GPU, I had some problems, I do not remember which. On the other hand, after installing an NVIDIA card and the NVIDIA FreeBSD driver, all worked perfectly.

My camera was a Logitech, I think a C920, and that worked OK, both video and audio on FreeBSD.

Yep, even documented it:

https://vermaden.wordpress.com/freebsd-desktop/

I'm using KDE and sddm. Last I heard, Zoom works for video but not sound.

FreeBSD and i3...and i don't touch Zoom, but my D810 works very well as a webcam ;)

Wow. Some people still care about PowerPC. Can't wait to run a fully functionnal FreeBSD desktop on that old IMac G5.

As nix23 alluded to, there's two sides to this. FreeBSD will run on the G5, but the real target of this port is newer POWER8 and POWER9 hardware, which is much more widely available than older IBM POWER gear, is modern, and is the fastest available platform you can run with no binary blobs in firmware, if that sort of thing bothers you. Or, if you just like the novelty.

True, he is probably better served with netbsd i think?

The last time I tried playing with older PowerPC Macs, I didn't have much luck with any of the BSDs. Linux's hardware support is just much better there.

Sorry?! I buyd for lots of money my talosII, and yes i really care about powerpc and freebsd.

This is a really nice write-up on using FreeBSD as a destop:

https://www.unixsheikh.com/tutorials/how-to-setup-freebsd-wi...

> Finished setting up the Malaysia mirror site, generously hosted by the Malaysian Research & Education Network. Traffic from Oceania and parts of Asia is now going to this mirror instead of farther away sites like Japan and California.

I see open source projects still listing a bunch of download mirrors in different countries - but this doesn't actually matter, surely, and I don't see how it even mattered in the 90s unless you're in Australia. Is there some technical reason they can't use anycast or similar load balancing?

I assume you aren’t questioning why there are globally distributed mirrors, but rather why they have to be explicitly listed, as opposed to automatically used. I think we can all agree that distributed mirrors are a good idea. They reduce load on undersea cables, can often offer more bandwidth, and have lower latency.

But, as for why they are listed... my guess is pretty simple: advertising. These are open source projects that don’t purchase space at these providers, the space is donated. The companies that give them space still need to get something in return. An open source project doesn’t have money, but they can offer eyeballs. So in addition to generally being good citizens, they can also get a little bit of good advertising.

You would be surprised how bad peering and back haul networks are still today. A normal business line here in the united states can still see sub 10Mbit/s going even out to hosters like OVH or Hetzner in France / Germany. For large projects like this, its always a good idea to have global mirrors if you can help it.

I wonder if you could build yourself a VPN to the nearest AWS/google region, and then uses that cloud's private fiber network to accelerate your internet speed to worldwide regions?

To the extent that you're willing to pay their data transfer fees.

That's what Cloudflare Warp does.

> and I don't see how it even mattered in the 90s unless you're in Australia

I remember choosing a nearby mirror from a list when downloading stuff in the early 00s in Europe, it seemed to make a significant difference.

Cost, complexity, not available from the provider who is donating their space/network

Maintaining an HTTP redirect load balancer does have issues, but it seems like if the mirrors have similar enough behavior (like they agree to all take the same URL path), you could run DNS anycast without involving them too much.

This seems better than asking the users to pick their preferred mirror (how do they know who's good?) and maintain it when the list changes. I wonder if anyone provides this as a service without the rest of the CDN.

Redundancy. Geo Redundancy.

I have a question: Let's say you compile FreeBSD yourself (or another kernel with 64-bit RISC-V support), and you decide that you don't want the C-extension (compressed instructions). Will you be able to switch this extension off without encountering any issues, for example with assembly? I have a feeling that you can't even write compressed instructions with assembly unless you hand-write the instruction opcodes yourself, so maybe it's not an issue?

I thought they were the same instructions, just a more compact representation (that has a subset of functionality)?

They are, but there are some drawbacks: 32-bit instructions can cross page-boundries and 16-bit instructions makes the length of the instruction dependent on a bit-pattern instead of a fixed-width.

Yes, they should have specified that even with the C extension, 32-bit instructions are 32-bit aligned. That would have slightly worsened density as you'd need a 16-bit nop to pad an uneven number of 16-bit instructions, but would have simplified decoding.

Reply on news.ycombinator.com