Systemd In Ten Years Has Redefined The Linux Landscape

[Paul Frederick]

Not at all you will lose the title of being a time wasting troll. Keep tittle of being a deceptive idiot by not providing correct information in the first place to make your problem look more important that what it really was. Because you hated systemd and don't tolerate change. Heck changes in consolekit before systemd broke lots of things.

It was important to me. When I built my PC I wanted the same setup only faster. That's not exactly what I ended up with though. But I'm making due I suppose. How in God's name did you become a senior member here? I've probably been running Linux for longer than you've been alive.

[oiaohm]

It was important to me. When I built my PC I wanted the same setup only faster. That's not exactly what I ended up with though. But I'm making due I suppose.

You still have not provided the information you should have. Other people time is important as well as you should not make them guess what was your problem just to win a point.

How in God's name did you become a senior member here? I've probably been running Linux for longer than you've been alive.

Really that absolutely not possible. I was one of the ones who build Linux first release on the Minix mailing list. So there is only a handful of people who have used Linux longer than me. By the way tittle Senior is kind of earned.

[Paul Frederick]

You still have not provided the information you should have. Other people time is important as well as you should not make them guess what was your problem just to win a point.


Really that absolutely not possible. I was one of the ones who build Linux first release on the Minix mailing list. So there is only a handful of people who have used Linux longer than me. By the way tittle Senior is kind of earned.

Well you sure don't come across as particularly knowledgeable.

[oiaohm]

Well you sure don't come across as particularly knowledgeable.

I appear exactly how I want to appear. Just because a person has decades of knowledge does not mean they just show it off. You still have to answered correct what was your issue in 2017. You say i was a DE from what I know I totally suspect not. Or it was a pure lie. This is the problem I have a lot knowledge I am not past the possibility of a mistake.

[Paul Frederick]

I appear exactly how I want to appear. Just because a person has decades of knowledge does not mean they just show it off. You still have to answered correct what was your issue in 2017. You say i was a DE from what I know I totally suspect not. Or it was a pure lie. This is the problem I have a lot knowledge I am not past the possibility of a mistake.

Indeed. My issue was I could not shut X Window down. Today I don't run a DM so it is not an issue anymore.

[k1e0x]

Yes not that can measure simple. The extra memory usage of Jails or cgroup is kernel based stuff normally not directly linked to process accountancy. This is where you logic is wrong as at times it can be way heavier than hardware base virtualization.



This is one of those horible thinks that people think is true but in reality is only sometime true. Yes the majority of cases it true but there are examples where it false..

https://indico.cern.ch/event/384358/contributions/909221/attachments/1170419/1689415/151014_hepix_wataru_takase.pdf

Page 11 here done with docker vs kvm there are old ones with solaris zones vs hypervisor I cannot find that were done by vmware that show the same kind of strangeness.

Notice the ping with docker is way faster than kvm but the sshd is way faster with kvm than docker. So one of those very warped things where a VM can be faster running 2 Linux instances or two freebsd instances than using cgroups/namespaces or jails

Its really horrible that KVM have such a wide performance profile with hardware acceleration is somewhere between the 0% to 80% performance loss . The advantage of docker, jails and zones you don't have this horrible wide profile normally 1 percent to 50% performance loss . So like it or not there are cases that docker, jails and zones lose vs kvm or other virtual machines.


But when it comes to process isolation due to the performance problems we see we know that FreeBSD and Solaris did not have the design right.


Kind of true.



This is kind of what you call R&D. Thinking at this stage we don't have a design that is perfect for this the R&D process still has it place.

Btrfs was in fact designed by oracle funded developer so both Btrfs and ZFS were fully designed implementation of both have run into major problems like ZFS absolutely horrible memory usage. https://lwn.net/Articles/747633/ XFS lead developer is taking a different route that may render both Btrfs and ZFS design ideas totally invalid. XFS one asks a really good question if the Unix old saying that everything is a file as true why have we not been able to just straight up use file system drivers on files. Why is snapshotting not designed around files as well.



The horrible problem is the solution may not be exactly engineering. It may be the total inverse.

Think about this you give 2 art classes just a block of clay no tools. One class turns in all bricks because the clay was baked solid before you gave it to them and only 1 student is in fact happy because they did not have to do anything because it was exactly what they wanted this is kind of the engineered solution you made sure you got the result you wanted. The other class get unbaked clay and you have huge lot of variation and most of the students are happy bar the 1 that did not want to do anything.

What would be the equal of clay. This kind of would be bpf in the Linux kernel so you can upload highly optimised configuration for what you are doing.

These problems are serous-ally not simple.

Well.. calling bs. For a while it almost sounded like you knew what you were talking about. Good luck with that evolving thing..

[oiaohm]

Indeed. My issue was I could not shut X Window down. Today I don't run a DM so it is not an issue anymore.

https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=861956

In 2017 we had the above bug. Yes X Windows would not shutdown if it was not active on the screen at the time. Interesting enough what happened here makes an absolutely unkillable program. So you are using systemd it prints out some errors and fails to shutdown quickly. sysvinit just locks up. This is not systemd bug. This is a X.org server major screw up triggering a kernel fault. So x.org bug and kernel bug leading to absolute disaster nothing todo with changes in systemd at the time. Last time systemd was documented as being the fault cause for a DM issue in 2013 all faults with DM since then have in fact been x.org X11 server related lots of them resulting in unkillable processes. .

This is another problem systemd informs user of problem instead of seeing hey systemd is telling them X item is broken. Somehow the user when informed by systemd takes the point of view that systemd has to be broken that is incorrect.

Yes there are times you hit problem that bad that no matter how good you code userspace service management there is no way in hell it can be solved at this point telling user what has gone wrong is the right move.

[oiaohm]

Well.. calling bs. For a while it almost sounded like you knew what you were talking about. Good luck with that evolving thing..

Before calling it bs do spend some time going through the benchmarks on the topic. I am interested to see how Redox version turns out and if it suffers from the problems that you can detect in jails, cgroups/namespaces and zones that result in them being slower than VM under particular workloads or do they find the solution.

The basic problem is that a lot of people think that isolation of processes is like a solved game like tic tac toe that is something you can engineer out a solution because you understand all the problem space. Isolating applications from the different benchmark failures it clear those have not found the perfect solution and that means every solution so far has missed something about the problem space.

Basically your ideas are right if we have successfully solved the problem space of how to isolate applications. If we have solved how to isolate applications properly zones, jails, cgroup/namespace... under any workload should be better than VM and minor-ally worse than bare metal. Problem is benchmarks don't tell this story. Instead the benchmarks tell the story that we have those systems at times worse than VM so we have not solved it yet.

While something is not properly solved how to-do it you cannot engineer a perfect solution either. Its like why you make a prototype and refine it before you engineer final product as well.

Of course someone could get lucky and engineer up the correct solution or the linux kernel chaos development model could try enough different things to point to where the correct solution is as well.

[timofonic]

I don't agree with this one. I'm using desktop system with systemd and utilizing heavily the user services for multiple things (including timers, services etc.) I actually start with "can this be a user service" before adding something as a system service. It works great and I've got nice statuses/logs & dependencies system to tie it all.

Can you please elaborate? I just read a lot of empty words in your message, it reads just like cheap Microsoft astroturfing :P

[Lizintacer]

Can you please elaborate? I just read a lot of empty words in your message, it reads just like cheap Microsoft astroturfing :P

I've got a restic service template that I parametrize with a couple of endpoints (AWS, B2, local). Then a cleanup timer that removes downloads 1 year older. Calendar service that synchronizes caldav. I run Sway as a systemd user service that actives a user target (sway-session.target) that in turn actives more graphical services (Mako, udiskie, redshift). I've got a service that's activated when one of these user services fail and shows notifications.

Is that what you wanted to hear?