Linux 4.3 vs. Liquorix 4.3 vs. Linux 4.4 Kernel Tests

[cip91]

Interesting, regarding games I dind't expect the minimum FPS to be lower than stock kernel. I wonder how the SteamOS kernel would compare.

[Azrael5]

Excellent: Improvements prevail on regressions.

[thelongdivider]

There are some latency micro-tests and such, but hard to represent latency tests in the real-world in a fully-automated / reproducible manner.

Good job Michael. I really appreciate these tests. It's understandable that as a 1 man force you don't have a lot of extra time for investigation, yet still you take requests from premium members for articles!

[Michael]

Perhaps Phoronix Test Suite does not understand latency at all...

There are some latency micro-tests and such, but hard to represent latency tests in the real-world in a fully-automated / reproducible manner.

[Michael]

A problem with these benchmark numbers and all others published here is that no attempt is made to understand what is happening. Active benchmarking would do wonders here:

http://www.brendangregg.com/activebenchmarking.html

It would be nice if the results had an explanation of why differences in numbers occur. I am not familiar enough with the 4.4 changes to guess on C-Ray, but I would guess that PostgreSQL does better because of BFQ. It would also be interesting if the remarks about debug kernels were examined. The "debug" features typically enabled in production do little to performance. The only thing that might matter is stripping on memory constrained systems, although if things go wrong, few tools can be used to analyze it. That includes things being slow. You need to compile a "debug build", but quite honestly, a "debug build" in software just means assertions are enabled. All other features that aid analysis do not matter for making a build a debug build.

Unfortunately given the ad-blocking situation, lack of other support, and just trying to make ends meet, it really isn't feasible doing the fully investigative part for articles like this where I am not the domain expert on all areas. The point is pushing out a lot of data, ensuring all of the data/benchmarks are reproducible, using relevant hardware, and that hopefully others interested in said results and with more domain expertise on particular area can share some insight or carry out their own research thanks to the data at hand, etc.

[ryao]

A problem with these benchmark numbers and all others published here is that no attempt is made to understand what is happening. Active benchmarking would do wonders here:

http://www.brendangregg.com/activebenchmarking.html

It would be nice if the results had an explanation of why differences in numbers occur. I am not familiar enough with the 4.4 changes to guess on C-Ray, but I would guess that PostgreSQL does better because of BFQ. It would also be interesting if the remarks about debug kernels were examined. The "debug" features typically enabled in production do little to performance. The only thing that might matter is stripping on memory constrained systems, although if things go wrong, few tools can be used to analyze it. That includes things being slow. You need to compile a "debug build", but quite honestly, a "debug build" in software just means assertions are enabled. All other features that aid analysis do not matter for making a build a debug build.

[GraysonPeddie]

Isn't the point of Liquorix and alike latency/responsiveness? which is not tested here...

Perhaps Phoronix Test Suite does not understand latency at all...

[geearf]

Isn't the point of Liquorix and alike latency/responsiveness? which is not tested here...

[xpris]

Is he still charging for it? Otherwise would be happy to run some tests of it.

Michael NeteXt'73 app is free. We can install it and use for free special optimized netext basic kernel (or use non optimized kernel from Ubuntu PPA - this gives us simply way to install it) also we can install optimized AMD or NVIDIA driver, also new firmware and microcode, APM and GCC. This gives my old laptop nice boost
If you want better choice you can pay for premium kernels - this should givers better performance but I still use basic kernels.

[GraysonPeddie]

A 4.4 generic kernel is faster than 4.3 Liquorix for C-Ray? Hopefully Liquorix could release a 4.4 kernel when it comes out and will be in AUR by then.