Artix Linux Forum

[Request] Performance optimized repositories (x86-64 v2/v3/v4)

30 July 2023, 13:12:12

There's plenty of attention around providing the ability to use packages compiled against LTO, PGO and higher architecture levels such as x86_64-v2, x86_64-v3 & x86_64-v4. ALHP and CachyOS has already proven that this yields greater performance in range of 10-36% in various scenarios. This is also being discussed in Arch Linux but currently the progress is halted due to limitations of their current infrastructure. Furthermore numerous systems are already providing LTO/PGO/x86-64v4 such as OpenSuSE Tumbleweed and soon Fedora 39 with LTO and x86-64 v2/v3/v4

Unfortunately ALHP and CachyOS repositories can't be easily used in Artix without manually removing packages which depends on SystemD.

Artix's biggest advantage is freedom of init system, decoupling from SystemD and power of rolling release.
This would be another major step forwards

Re: [Request] Performance optimized repositories (x86-64 v2/v3/v4)

Reply #1 – 30 July 2023, 19:58:44

I don't know exactly what was done, but for a while Waterfox moved to v-whatever builds, and it wouldn't run on older core 2 duo cpu's. But then they stated in later update notes this limitation had been removed because they found some other compiler optimisation flags that gave the same performance increase without doing it like that.

Re: [Request] Performance optimized repositories (x86-64 v2/v3/v4)

Reply #2 – 30 July 2023, 20:20:07

Yeah indeed, so the thing is, on average, the performance/efficiency benefits add up little by little:
compiler flags (what you mentioned above): 5%
forcing architecture to v3: 5%
forcing specific cpu architecture class (reason why ppl use gentoo/lfs): another 5%
...so on and so forth

I do use v3 repos myself and the cachyos-v3 kernel which has probably the best performance possible, but for a smallish distro like here i'm not sure the server space is worth wasting on duplicates of all packages as unoptimized/v2/v3. I might be wrong of course. Arch upstream has been discussing this for a while but they didn't implement it yet: https://gitlab.archlinux.org/archlinux/rfcs/-/merge_requests/2?commit_id=1a532bcfc37e280bf69b219bb44308d863dee476

Maybe what Artix can provide right now, and it'd be really cool, is just kernel and browsers/java/other heavy stuff in -v3 variants. I would certainly like that.

Re: [Request] Performance optimized repositories (x86-64 v2/v3/v4)

Reply #3 – 31 July 2023, 09:59:23

I find it really amusing, the same people who replaced KISS with "mainly new and because it's safe" are suddenly talking about performance?

Re: [Request] Performance optimized repositories (x86-64 v2/v3/v4)

Reply #4 – 31 July 2023, 14:19:02

You mean Arch? Afaik they always talked about that. Gladly they still take server users' stability concerns into consideration, of course you wouldn't use unneeded flags there, maybe just architecture optimizations.

Re: [Request] Performance optimized repositories (x86-64 v2/v3/v4)

Reply #5 – 27 September 2023, 10:23:24

> This would be another major step forwards
All I see is a step back. Change the computer for the mantainers' or manufacturers' wants to change something in the architecture? Give everyone money for it, then we'll talk, otherwise only owners of such architectures will sit for such distributions.

Re: [Request] Performance optimized repositories (x86-64 v2/v3/v4)

Reply #6 – 27 September 2023, 16:13:03

@TerminalHash If upstream Arch reaches an agreement over this, be assured the repos will be split probably in two to save space, between generic (normal as usual) and v3 (as v2 has too little benefit and v4 almost noone uses).
Some kernel devs were complaining when even the i486 from 1989 was about to be dropped, imagine with any 64bit architecture, lol.