Jump to content

Recommended Posts

Posted

it seems to be only for a memory-remap function and not for the whole system ;)

Quote

Back in 2018 was a memory management speed-up by around 20x for the mremap system call on large memory regions. 

 

That patch by a Google engineer was an optimization for mremap given that Android relies on using it for large regions of memory during various operations. The mremap system call can be quite slow without transparent huge-pages (THP) while this patch makes things faster by copying at the PMD level when possible.

 

The speed-up back in 2018 by the engineer was reported to be around ~20x faster on x86 (x86_64) with a 1GB mremap taking just 144~160 microseconds rather than 3.4~3.6 milliseconds.

 

But for systems with THP support, there isn't likely to be much of a performance difference.

 

 

Posted

this improvement only in some certain cases. and the cases are mixed in our daily use, maybe not high frequently used, so 8x~20x * frequency, that's actual improvement, we can feel.

Posted

It does seem that the optimization is for certain use-cases only, not apparent in daily desktop operations. Anyway, 20201123 is running without issues so far, congratulations to balbes :thumbup:

Posted

eth0 seems to disengage after a while. Reverted to kernel 5.9.9, seems stable so far. Looking forward to 5.10 final after this weekend.

Posted
10 minutes ago, amirul said:

eth0 seems to disengage after a while.

What box/sbc are you seeing the eth0 issues?  I am seeing something similar on a N6 max (rk3399) but don't have issues with a TX3 mini (s905w) when running 5.10.0-rc7.

Posted

It is my own kernel build from mainline kernel sources.  I build 5.4 and 5.9 kernels for my 'production' boxes.  I just this past weekend got a build environment put together for 5.10 for my test boxes.

Posted

I will let you know.  One of the changes between 5.9 and 5.10 for the rock-pi-4 dtb is that in 5.10 it now has three versions rock-pi-4a, 4b and 4c.  I am currently using the 4b version for the N6 max.  If the networking stability remains, I may experiment with the other dtb files and/or try to revert back to the way they were structured in 5.9 to see if that may be the cause of the issues. 

 

I bought the N6 max board because of your recommendation and am very impressed with the overall performance.  But I really don't know how closely the box hardware matches the rock-pi-4 dtb that I am currently using.  This is the most recent board I have purchased and I am still getting to know it better.

 

Do you run your N6 max box from sd card or internal emmc?  It took me a bit of work to adjust balbes scripts, but I eventually got the os installed on emmc.

Posted

I'm running off sd and since it is running well enough, I didn't to risk bricking the box by trying to install to emmc. What cpu speeds have you got running?

Posted

I am running at 1800.  I haven't looked into trying to increase that any higher.  As this is my newest box that I am still feeling my way around, I'm not running anything on it that really pushes the cpu, except for a benchmark utility here or there.

Posted

I've got it running the xfce4 desktop at 1.70Ghz(cpu1-4) and 2.02Ghz(cpu5-6) plus a few other tweaks. With kernel 5.9.9 the issues with eth0 seems to have gone, although there were a few times it disconnected/reconnected on its own.

Now that 5.10 final is released, waiting for Armbian :)

Posted
4 hours ago, Bozza said:

Is there a comprehensive list of changes made that are more of an interest to us arm users? 

 

Armbian supported hardware or in general? We usually integrate features (way) before they reach mainline.  Perhaps quick scan here https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/log/arch/arm64?h=v5.10.1 and here https://www.collabora.com/news-and-blog/news-and-events/kernel-5.10-rockchip-h264-bifrost-more.html ?

Guest
This topic is now closed to further replies.
×
×
  • Create New...

Important Information

Terms of Use - Privacy Policy - Guidelines