performance differences in Linux distros?

[HugoNotte]

Since it has been established that FAH performs a little better on Linux than on Windows, I'm wondering whether anybody has observed differences between different Linux distributions?
The guys at phoronix.com have done several comparisons between different distros and some seem to be doing better than others, depending on the type of workload: https://www.phoronix.com/vr.php?view=29139

Seems like Clear Linux from Intel is quite fast throughout, but might be a bit more for geeks. By reading on this forum it looks like that a lot of FAH is run on Ubuntu / Mint, which might not be the fastest base.

Has anybody made comparisons specifically for FAH?

[NRT_AntiKytherA]

The only difference I have personally observed is how easy or not it is to setup on different distributions. I found Arch based Manjaro easier to install fahclient and fahcontrol to than I did Ubuntu 20.04 which required some faffing about. Apparently there will be a new Linux fahcontrol released as and when there is time in the future which should resolve a lot of the dependency issues though.

Actual performance will reflect the phoronix testing quite closely. Clear Linux is used by AMD to benchmark their new hardware, kind ironic considering Clear Linux is an Intel project but it is testament to the fact it is devoid of detritus.

[MeeLee]

Clear linux doesn't have any installation candidate.
You'll have to compile fah client for the operating system; so unless you know what you're doing...
Also, clear Linux performs better at 'a lot of' tasks, but does still perform worse than Ubuntu in some tasks. It's kind of a mishmash, and no one knows what program will work better or worse before actually trying it out.
The benefit clear linux currently has over many other operating systems, is that it is a small and simple operating system; meaning less overhead.
Though Linux is pretty good.
No one has actually tried a side by side comparison of MacOs, Debian based Linux, Redhat based Linux, Clear Linux, or Unix.
MacOs being a derivative of Unix, in theory, the client could also work on it, if compiled correctly.

Ubuntu and Mint, they all have the same underlying Linux core (the engine of the OS).
Just they are more the luxury sedans, while Lubuntu is like an economy car.
All of them you can run headless, basically run the engine on a bare bones platform.
Personally I run the GUI, but in theory it's possible to log out, press CTRL + ALT + F1 (or F2), and after logging in, type:
sudo init 3
sudo init 5, to reverse the command and restart the GUI.

You should have pretty close to no system resources allocated to the GUI that way; and run pretty close to a headless version of Linux.

[foldy]

Can FAH run easily on Clear Linux? It seems 5-10% faster in general than Ubuntu

[HugoNotte]

No one has actually tried a side by side comparison of MacOs, Debian based Linux, Redhat based Linux, Clear Linux, or Unix.
MacOs being a derivative of Unix, in theory, the client could also work on it, if compiled correctly.

And that was my question, whether anyone so far has made any observation, that one Linux distro tends to allow for more PPD than another. I would have thought that Arch Linux should be pretty good, but looking at the benchmark results for Manjaro, it might not be any better than Ubuntu.

I personally run Mint 19.3 and know that the kernel is quite fat. But since usability is a factor for me, I would be hesitant to change to another, less user friendly distro just for the sake of 500 or 1000 more ppd.
I guess, if one wants to be really hardcore, one could compile one's own kernel specially for the individual machine it runs on and start from there.

Just by the way, MacOS is actually Unix not just a theoretical derivate, it does fully comply with the Single UNIX Specification (SUS). And the MacOS client does run very well

Anyway, was just a matter of interest question.

[bruce]

I think you're asking the wrong question. The performance doesn't depend on FAHClient; it depends on FAHCore_xx. For the most part, FAHCore_xx build contain statically linked library entries rather than dynamically linked componenets specifically to avoid inter-distro differences. Calls to I/O routines or driver routines, for example, are external links of course but it's unlikely that you'll see distro dependent speed differences other than things like NFS support code that's more dependent on your installation than on anything else.

[bikeaddict]

Running ldd on the FahCore_22 binary under Fedora 32 shows it uses a number of dynamically linked libraries. But how much WU processing actually makes calls to them is unknown.

Code: Select all

linux-vdso.so.1 (0x00007fff9faef000)
libOpenCL.so.1 => /lib64/libOpenCL.so.1 (0x00007f0006fb6000)
libpthread.so.0 => /lib64/libpthread.so.0 (0x00007f0006f94000)
libdl.so.2 => /lib64/libdl.so.2 (0x00007f0006f8d000)
librt.so.1 => /lib64/librt.so.1 (0x00007f0006f82000)
libstdc++.so.6 => /lib64/libstdc++.so.6 (0x00007f0006d92000)
libm.so.6 => /lib64/libm.so.6 (0x00007f0006c4c000)
libgcc_s.so.1 => /lib64/libgcc_s.so.1 (0x00007f0006c2f000)
libc.so.6 => /lib64/libc.so.6 (0x00007f0006a65000)
/lib64/ld-linux-x86-64.so.2 (0x00007f0006ffa000)

[MeeLee]

Since Debian and redhat based versions are most commonly used for fah,, it is safe to presume their cores also work the best.
Mac (and therefore also Unix) only support CPU folding.
Clear Linux doesn't have an installation candidate, so compatibility for GPU folding is questionable if it'll work at best.
If Clear Linux does CPU folding a bit better than Ubuntu, or redhat, will it be worth the struggle of getting it to work?
We're probably talking about 10%, if anything at all... no more...

[NRT_AntiKytherA]

The arch version is a repackage of the debian files, it works no faster on manjaro but may with arch linux as they are different beasts.

manjaro will run the arch linux fah packages BUT while originally based on arch should not be considered as arch linux with a more user friendly install routine and different theme. Some packages are still sourced from the Arch repository but that's mainly optional desktop environment packages and shared libraries rather than core packages. For example the manjaro developers compile their own optimised kernels from kernel.org source code and hardware drivers also from source.

Then there are all the automated hardware driver install routines, extra GUI applications written in-house.

clear linux runs all applications containerised and has support for flatpak but not snap (yet). If there were a flatpak you could get fah up and running fast, at present I know of a snap only though.

[MeeLee]

FAH only has 2 Linux distributions of their software.
For Redhat (.rpm), and Debian (.deb).
Mint and *buntu versions are debian based.
CentOs, Fedora are Redhat based (RPM).
Any other version of Linux either bases itself on one of these two, or uses a recompiled version; which means it's not particularly made for that operating system.

[NRT_AntiKytherA]

FAH only has 2 Linux distributions of their software.

Any other version of Linux either bases itself on one of these two, or uses a recompiled version; which means it's not particularly made for that operating system.

well aware of that, it says so in my post about arch and manjaro. the fah snap someone built opens up the compatibility a little for other distributions that actively support snap store too but it's the old client still

[MeeLee]

Forgot to mention the MAC version, which is Unix, similar to Linux.
What I meant to say, is that more than likely, performance won't be any faster on those other Operating systems.