Hi,
I stubled upon https://liquorix.net/ from the time I was running ubuntu, is there such a thing for Suse?
Regards,
You could install kernel-sources and then compile your own kernel. That way, you get to choose your own optimizations.
Makes sens.
I was hoping there is a version out there already compiled and optimized for newer hardware. I have a i7-6700HQ processor.
Tumbleweed usually has close to the latest kernel.
Reading Liuorix supposed optimizations,
Am amused at its continued suggestion that network “buffer bloat” is best addressed by shrinking buffer sizes.
I was one of the first to publicly denigrate the idea (at that time) that TCP/IP buffers should be returned to the sizes of WinXP (At that time, various experimentation revealed less latency with smaller buffers).
In my relatively long published article
https://sites.google.com/site/4techsecrets/optimize-and-fix-your-network-connection
I described how this was not a matter of buffer size but that when the buffer size was large it was being mis-managed.
After a few years of online back and forth (at times very emotional because after all, I was decapitating what some came to feel would be their most significant career achievement), nearly all including the original founders of the idea caved and started publishing new research that supported my line of thinking.
And, I still from time to time run into pockets of people who still believe in “buffer bloat.”
The article I posted above is still important for anyone who is deploying their openSUSE which is <not> <all> of the below
- Desktop of very small server use. Only relatively few simultaneous connections.
- Wired Fast Ethernet connections (802.3) up to about 100Mbits/sec.
- Non-busy networks, limited or no network congestion.
- A completely clean network free from various interference. Possible sources might include mis-configuration, EMF like elevators and microwaves, failing hardware.
Some other things of note in the liquoris kernel…
I see a number of kernel modifications which requires looking into, to determine whether they are really beneficial.
For example, would that “Budget Fair Queue” be better, worse or no diff compared to existing disk schedulers?
Would “hard kernel preemption” have a downside under lesser loads?
In my research on TCP/IP Congestion Control, Linux has many choices but none that “fit all.” The YeAH Congestion Control Algorithm should be understood well before implementing (see my article above to understand)
I don’t know the current mainline kernel status of AUFS, but various UFS are already supported.
If there is a real lesson to be learned, IMO it’s to be wary of optimizations and to implement them only when relevant. Unless your use and scenarios are relevant to say… the specific optimizations in the liquoris kernel, avoid it. It’s unlikely that someone’s customizations would be relevant to the general population as a whole although may be important to specific persons.
Perhaps the other important lesson to be learned is that unless you want/need a specific unavailable kernel component or setting, you can do a lot with today’s modular kernel architecture. My article describes how to set TCP/IP network buffers and the Congestion Control Algorithm to something different. In a different article, I’ve written how to set the disk schedular to optimize for SSD. And, there are numerous other ways you can poke and set /proc to different values which don’t require a whole new kernel.
IMO,
TSU