Kernel 4.1 incredibly fast!!

[dotlj]

The 4.1.0-1-amd64 kernel in my Debian testing vms boots in around 13.2 seconds, mean.
The 3.16.0-4-amd64 kernel in my Debian stable vms boots in between 12.0 and 13.0 seconds, with more variation.


The 4.1 perhaps seems faster, but on my old hardware from 2010, not much difference. YMMV.
The 3.16. can be about one second faster. Both kernels have not been customized.
Boot time for me is not important because I only reboot after kernel upgrades.
Host hardware intel i5 M20 running Debian stable. Both vms have same single core and same memory.

[spacex]

@dotlj

Yes, but it's not comparable unless you try both kernels IN THE SAME release, with the same hardware. There are other differences between stable and testing besides the kernel, so it isn't comparable.

[runfrodorun]

I call shenanigans -- post some comparative objective benchmarks please.

In case you hadn't noticed, this OP is 100% opinion, 100% of the time.

I kept encouraging him to join the systemd "debate" threads; he's perfect for them. But nooooo.

If every opinion was valid there might be more windows users around these parts

-RJ

[Wheelerof4te]

Just installed 4.1 kernel from jessie-backports. There may be some improvement in overall performance, but that is hardly a surprise. Linux kernel is an open-source software, meaning it will get better as newer versions come out.

It also means there will be nasty bugs that can suddenly crash your system, or simply annoy you. You do the trade off, as with any new, untested software. That's why Debian waits a while to accept things into backports.

So, my point is: You shouldn't be so hyped when new, better things get tested and it just works.

[spacex]

Just installed 4.1 kernel from jessie-backports. There may be some improvement in overall performance, but that is hardly a surprise. Linux kernel is an open-source software, meaning it will get better as newer versions come out.

That's NOT always true for older hardware.

[Wheelerof4te]

^My hardware is quite old (graphic card Nvidia Geforce 4 from 2002.) CPU is maybe even older, but with PAE. Still, machine runs great and i don't see any difficulties with this kernel.

Now, it would be different story if you referred to even older PC's. That would mean ANCIENT, not old.

[Head_on_a_Stick]

@spacex is right, hardware regressions are not uncommon in Linux.

For example: https://bbs.archlinux.org/viewtopic.php?id=196116

[bester69]

My pc is an Acer Extensa from 2009 with just 1GB RAM and 1GB CPU, and

i felt the performance improvement when upgrated from 3.16 to 4.1 kernel.

I dont know what brigns 4.1 that in my case improved performance IO disk, but the following article i think it says something like that.:

Some of our highlights for the Linux 4.1 kernel include:
- Significant performance improvements for certain hardware as well as power consumption/efficiency improvements for select Intel hardware.
(http://www.phoronix.com/scan.php?page=n ... ature-view)

[spacex]

I've only tried kernel 4.1 in Sid, which isn't comparable with trying it in Jessie. But what I found, was that Sid used more RAM than Jessie.Just slightly, and I have a lot of RAM to spare. But still, me don't like to use RAM

[runfrodorun]

Yeah, for older hardware, some things run better in kernel 2.4 (back in the day). Consider there is always a reason, and usually a darn good one to abandon old methods.

Consider- methods of video accel have changed so much in the last 15 years, that the technologies from the early 2000's and late 90's are like a world away in terms of what is used for video acceleration. When programs like firefox are updated to take advantage of the latest and greatest accel methods, while discontinuing support for these old methods as they are hardly even relevant to the kind of operations that modern software wants in order to achieve the desired result (which is looking pretty usually, totally different rant). And if supporting multiple accel methods over the course of 15 years was so easy, maybe Mac OS 10.6+ would run on the G5 (Spoiler: it doesn't), cpu architecture differences sort of being the analog here.

So yeah, things can get pretty darn slow, or at least, if you're looking at 'benchmarks' or 'performance ratings' as an all encompassing number to order systems by, then what you see won't match up. A graphics card may be rated as 'half as fast' on some benchmarking utility, but you might observe 1/100 the performance based on driver support and accel methods used.

Disclaimer: Very non-technical explanation here and I'm aware of that.

-RJ

[spacex]

But Debian does support different accel methods. At least for Intel graphics. Just because something is default, it doesn't mean that the other methods aren't supported...

[Chiefahol]

A dogmatic scepticism which seeks to immediately shut down any further discussion or comparison is just as stubborn as pure confirmation bias.

That's the smartest thing i've read all day!

[runfrodorun]

But Debian does support different accel methods. At least for Intel graphics. Just because something is default, it doesn't mean that the other methods aren't supported...

That is absolutely true, but that doesn't mean they are utilized. If there are better ways to implement some sort of visual effect that are available on new devices and drivers that are either not available on old devices and old drivers or need to be emulated with old devices and new drivers, then one of two things happens:

1. There is an abstraction layer in some library, say openGL for example that makes this invisible to software developers. If this is the case, then things might work, and may be as fast on older devices, but often times will not be because the transformations necessary are often times not optimized, because the principal behind the accel method is not isomorphic to the source of the transformation.
2. It is purely up to the willpower of the developer to maintain the ability to use that accel method, and I think we know how likely it is that they will continue to do so. I sure wouldn't! too much work

-RJ