@ddg Mainly that if you're in a highly congested network that also respects DSCP, interactive SSH might be slower/laggier.
You could set it to "IPQoS af21 none" so interactive sessions keep the previous setting, but even that setting supposedly causes worse problems in Japan (for me af21 behaves as I described, but other people are saying it doesn't work at all and packets are dropped).
Honestly though, anything but "none" is probably not very useful unless you're on a known network that handles non-default settings better.
Thank you for the explanation! I don't usually need more than 10mbps, but I'll leave the "IPQoS none" setting commented on my config file just in case I ever need more.
There's no chance I'd ever be able to find this solution myself, even if I were to notice the connection should be faster… Thank you again!
@mirabilos I don't think that does anything for rsync, as it shouldn't be allocating a tty anyway by default.
Both default QoS settings are wrong in Japan (interactive and non-interactive). For me they're both slow, but for other people the interactive setting doesn't work at all (packets dropped). You really need to override it with IPQoS (or mangle the packets at the router).
If anyone's wondering how I figured this out... I was wondering why v6 SSH is slow and v4 is fast. I thought it was a routing/ISP issue with v6 somewhere, but then I tried netcat and it was fast too.
I was wondering if it was DPI or something, but when I tried bouncing SSH through socat on the *client* side that was fast too.
I ended up using strace on socat and SSH to see what they did differently, and that's when I noticed ssh sets DSCP!
@iyes The only reason it doesn't happen with IPv4 is that is tunneled over IPv6 and the tunnel doesn't transfer the DSCP setting to the outer packet headers... so I'm using IPv6 anyway.
I had... not thought to benchmark, I just assumed that the data centers I was connecting to were rate limiting or adding some kind of latency (DigitalOcean mostly).
I'd only do this on a host by host basis. Normally low latency is more important and 10 Mbps is plenty. Only sites you rsync or git clone from need it changed.
I wonder if they intend that to apply to customer traffic or if that QoS rule is meant for their management plane and is being erroneously applied to customers, reserving 10mbit for ssh to manage equipment makes sense and is fine for cli access but not bulk data. Now I wonder if ssh can set a different flag when doing scp/SFTP or by setting ssh options in rsync/git, if the QoS options help interactive performance on a loaded connection
You can pass two values to the option to set it differently for interactive vs non-interactive traffic. I'm not sure it's worth it though...
They do have a tech document saying that DSCP 8 (cs1, the SSH default for bulk traffic) is considered priority, though they don't mention the bandwidth cap. So it think it's intentional.
I get ~3ms ping to Google, so it's definitely not laggy. Also the default interactive DSCP doesn't even work for some users (the packets get dropped), so this behavior is not intended to work well with SSH defaults, it's just a coincidence that the SSH default (which changed at some point) causes this ISP behavior.
SSH uses AF21 for interactive and CS1 for non-interactive. CS1 is intended to be lower priority than the default, not higher. However, NTT IPoE only documents the CS1 value to be mapped to high priority, and does not say anything about any other value (and does not guarantee packets are forwarded for anything else). So it's actually all backwards.
“The big vision is to move past the reliance on any single country’s investment in this core life sciences infrastructure: It’s not to have a duplicative, parallel structure in Germany, nor to simply transplant a centralized system from the US to Germany. The goal is a fully #OpenSource, #federated, safety net, embedded within the international community, with a strong global network of support.”
I want to be really clear: I’ve never targeted anyone. The only times I’ve spoken up were when I felt I had no choice but to defend myself from Luna’s ongoing harassment and abuse.
Even now, I’m only responding because I feel deeply misrepresented and hurt by what’s being said about me.
I don’t want conflict. I don’t thrive on drama. I don't "go after" anyone. I just want to feel safe and respected.
If you’re not engaging in harmful behavior (like harassment and doxxing) then truly, you have nothing to be concerned about. I don’t hold grudges or act out of malice.
That said, if someone chooses to publicly support her while she continues to harm me, I may need to quietly create distance for my own emotional safety.
I hope they can understand that this isn’t about punishment, it’s about protecting myself.
@gamingonlinux Crazy question. You mention Steam should denote games with Denuvo activation. Is it noted on Steam before you buy the game, or only after it is purchased? gamingonlinux.com/2025/05/denu…
Although for some of us this is a well-known problem, it has appeared again with the release of DOOM: The Dark Ages so it's worth a quick PSA to remind people about Denuvo activation limits with Proton on Linux, SteamOS / Steam Deck.
thanks, I guess I've just never run into the issue. I ask because I usually run my account on 6 or 7 machines (work/home/mobile) and I'm always adjusting proton.
Valve updated Proton Experimental, the testing area for the latest fixes to their Windows compatibility layer to run more games on Linux / SteamOS and Steam Deck. Here's all the changes for the May 16th update.
The Wine 10.8 development release is out now for the Windows compatibility layer bringing with it various bug fixes, along with user handles in shared memory for better performance.
(I'm not going to pretend I know exactly what this does, but it seems like a neat example of Linux being used in music production that we hope is helpful to musicians in the community 😄)
main difficulty of linux usage in music making is DAWs and plugins. Plugins work better in linux thanks to the wine guys being amazing but running a industrial program on a vm just dont make sense.
Other stuff is they are closed source.
Presonus and some of others are trying to build their daws on linux, there is bitwig too. But i would love a usable daw, which is open source in linux.
Plugins are hard because license stuff but thanks to pipewire and kernel i would love a good daw
about 10 years ago, I had a friend who bought a cheap midi keyboard not realizing it had no actual hardware inside, so he needed a PC to go with it. He was broke, so I dug out some ancient motherboards and cases from my closet and frankensteined him together a linux box and installed Ubuntu Studio on it. Had a spare USB->Midi cable, and boom. He had a working keyboard without spending any extra money. Linux is so great for this stuff.
📨 Can PostgreSQL replace traditional message queues?
Watch Joobi S B explores how #PostgreSQL can take on messaging workloads—simplifying architecture, improving reliability, and reducing tool overhead at #FOSSASIASummit2025
Auf YouTube findest du die angesagtesten Videos und Tracks. Außerdem kannst du eigene Inhalte hochladen und mit Freunden oder gleich der ganzen Welt teilen.
The Team 17: Chains of Command Humble Bundle is live with another pretty great set of games you can get for cheaps. Here's your usual expected run over compatibility across Linux / SteamOS and Steam Deck.
We started getting some random races in CI libtool: compile: powerpc64le-unknown-linux-gnu-gcc -DHAVE_CONFIG_H -I. -I../../datrie -I.. -I../.. -O2 -pipe -g -Wp,-D_FORTIFY_SOURCE=3 -Wp,-D_GLIBCXX_AS...
The crazy coders at Valve had been making the monthly top lists manually this whole time, but no more. They've given their monthly top-releases charts a fresh makeover and you can scroll all the way back to the release of Half-Life 2.
Here's the fresh and up to date list of games available on Amazon Prime Gaming for May 16th, and what compatibility you can expect for Linux / SteamOS and Steam Deck.
Ayame, a top-notch assassin who has lived 25 years without anything such as romance, is convinced by a colleague to register on a dating app, while hiding her true identity! She ends up matching with none other than her next target: Murai, the ruthle…
ddg
in reply to Asahi Lina (朝日リナ) // nullptr::live • • •Asahi Lina (朝日リナ) // nullptr::live
in reply to ddg • • •@ddg Mainly that if you're in a highly congested network that also respects DSCP, interactive SSH might be slower/laggier.
You could set it to "IPQoS af21 none" so interactive sessions keep the previous setting, but even that setting supposedly causes worse problems in Japan (for me af21 behaves as I described, but other people are saying it doesn't work at all and packets are dropped).
Honestly though, anything but "none" is probably not very useful unless you're on a known network that handles non-default settings better.
ddg
in reply to Asahi Lina (朝日リナ) // nullptr::live • • •Thank you for the explanation! I don't usually need more than 10mbps, but I'll leave the "IPQoS none" setting commented on my config file just in case I ever need more.
There's no chance I'd ever be able to find this solution myself, even if I were to notice the connection should be faster… Thank you again!
Seb 🇺🇦
in reply to Asahi Lina (朝日リナ) // nullptr::live • • •mirabilos
in reply to Asahi Lina (朝日リナ) // nullptr::live • • •--with-rsh='ssh -T'should switch to the bulk classAsahi Lina (朝日リナ) // nullptr::live
in reply to mirabilos • • •@mirabilos I don't think that does anything for rsync, as it shouldn't be allocating a tty anyway by default.
Both default QoS settings are wrong in Japan (interactive and non-interactive). For me they're both slow, but for other people the interactive setting doesn't work at all (packets dropped). You really need to override it with IPQoS (or mangle the packets at the router).
🗦new🗧 FireFly post-WHY2025
in reply to Asahi Lina (朝日リナ) // nullptr::live • • •Asahi Lina (朝日リナ) // nullptr::live
in reply to Asahi Lina (朝日リナ) // nullptr::live • • •If anyone's wondering how I figured this out... I was wondering why v6 SSH is slow and v4 is fast. I thought it was a routing/ISP issue with v6 somewhere, but then I tried netcat and it was fast too.
I was wondering if it was DPI or something, but when I tried bouncing SSH through socat on the *client* side that was fast too.
I ended up using strace on socat and SSH to see what they did differently, and that's when I noticed ssh sets DSCP!
ida iyes inoine
in reply to Asahi Lina (朝日リナ) // nullptr::live • • •Asahi Lina (朝日リナ) // nullptr::live
in reply to ida iyes inoine • • •abstrm
in reply to Asahi Lina (朝日リナ) // nullptr::live • • •@iyes
Is that DS-Lite they're using?
Asahi Lina (朝日リナ) // nullptr::live
in reply to abstrm • • •abstrm
in reply to Asahi Lina (朝日リナ) // nullptr::live • • •@iyes
Cool!
Jeremy 🇨🇦
in reply to Asahi Lina (朝日リナ) // nullptr::live • • •this made ssh snappier and I'm in the US.
I had... not thought to benchmark, I just assumed that the data centers I was connecting to were rate limiting or adding some kind of latency (DigitalOcean mostly).
Cassandrich
in reply to Asahi Lina (朝日リナ) // nullptr::live • • •Asahi Lina (朝日リナ) // nullptr::live
in reply to Cassandrich • • •Raven667
in reply to Asahi Lina (朝日リナ) // nullptr::live • • •Asahi Lina (朝日リナ) // nullptr::live
in reply to Raven667 • • •You can pass two values to the option to set it differently for interactive vs non-interactive traffic. I'm not sure it's worth it though...
They do have a tech document saying that DSCP 8 (cs1, the SSH default for bulk traffic) is considered priority, though they don't mention the bandwidth cap. So it think it's intentional.
training_solo
in reply to Asahi Lina (朝日リナ) // nullptr::live • • •Asahi Lina (朝日リナ) // nullptr::live
in reply to training_solo • • •I get ~3ms ping to Google, so it's definitely not laggy. Also the default interactive DSCP doesn't even work for some users (the packets get dropped), so this behavior is not intended to work well with SSH defaults, it's just a coincidence that the SSH default (which changed at some point) causes this ISP behavior.
SSH uses AF21 for interactive and CS1 for non-interactive. CS1 is intended to be lower priority than the default, not higher. However, NTT IPoE only documents the CS1 value to be mapped to high priority, and does not say anything about any other value (and does not guarantee packets are forwarded for anything else). So it's actually all backwards.
"Musty Bits" McGee
in reply to Asahi Lina (朝日リナ) // nullptr::live • • •Asahi Lina (朝日リナ) // nullptr::live
in reply to "Musty Bits" McGee • • •