Linux 5.15 Is A Very Exciting Kernel For AMD

[skeevy420]

Use ZFS for raid?

That's what I do.

My "multimedia" mirror started as a single 2TB disk back in 2015 two computers ago (Intel q6600)...split 75% ZFS and 25% NTFS...no shit... Went to 100% ZFS around 2016/2017, also changed motherboards (HP to an ASUS). Upgraded to a new (to me) PC about a year after that (Dell T5500, dual x5687) and moved my disk into that.

It then became a mirror on that PC in 2020/end of 2019, I attached a 4TB disk to my 2TB pool. Was such a tease having 2TB I couldn't access. A few months later I finally replaced the original disk when I built my current computer (custom R5 4650G build) and attached a 2nd 4TB disk to it.

I feel like all of that would have been a severe pain in the ass to do with any other file system or setup. I love being able to buy disks as I can afford them or as shit happens and it becomes necessary. I've gone through multiple motherboards, CPUs, operating systems, ZFS versions and features, and I'm not even using the same disk as when I started, but all of my important data and games have come with me.

If that's not a ZFS success story I don't know what is.

[mdedetrich]

Use ZFS for raid?

Its kinda hard to not use ZFS for RAID, I mean its specifically designed for RAID setups.

[edwaleni]

The last kernel where Michael got this excited was back at 3.18. That one even made my Athlon laptop run better and I installed it on all my Linux platforms.

He was right, it was a worthy release. So I tend to believe him when he gets excited about a new release.

[Jabberwocky]

I have 5 graphics cards, 2 laptops, 1 workstation, and 2 servers. All AMD and none work 100%. AMD just doesn't work that well in my experience.

I was wondering what you meant about "they need so many changes just to make their stuff work". I thought it was more severe, but understand now thanks. AMD does not dedicated the same amount of resources to Linux compared to Windows and I agree that shows. I would like it if AMD improved it (from the news here they hopefully will), but I don't feel like I can complain about it in my exeperience. Perhaps I just managed to get he parts that worked better for my use-cases. Intel has a better track record for implementing the entire product with advanded power control, sensors, and even odd communication protocols.

In the past 5 years I have bought/built following AMD systems: 3 laptops, 7 desktops/workstations, 5 graphics cards and 1 server. I had 2 big problems related to early adopting. I had to RMA 1 of the CPUs (1800X) and I struggled a few days to get my server 100% stable due to AGESA-RAM-woes. Besides those 2 release day purchase issues I had no problems*.

*I have some other problems but I don't think AMD is to blame. One of the CPU temps is not working. The problem is with the ASRock motherboard firmware, not AMD: https://github.com/rigred/sensor_config/issues/3 I have similar other issues with ASRock firmware... The server's GPU caused crashes frequently: https://forum.proxmox.com/threads/pv...ponding.35930/
I stayed away from any GPU after Polaris so experience is lacking in that area. I have borrowed some cards here and there, but no long term use. I know many people had CPU temp problems from the start, it's just by chance that my problems were ASRock issues and not CPU offset issues like others had. Still it's not like the CPU did not work or that you couldn't read other CPU temps.

My most recient Intel challenge was helping a friend has a Gigabyte Aorus 15P laptop, i7-11800H with a Geforce 3060 which retails for just over 2300 USD here, which does not boot into Linux. We tried various things but spent hours on it and didn't get it working. Laptops are usually the place where I find the most problems. I bought x2 Dell Ryzen laptops on the release day which worked well for me. Intel is probably not to blame for this, but we still have to figure it out... you win some you lose some.

To be honest when you compare AMD to Intel in this specific area historically there is a world of difference. That seems to be changing lately though with AMD stepping up their Linux support a lot in the past 2 years.

I agree with you, just from reading Phoronix I've seen how far in advance Intel commits support/features for unreleased hardware. I trust that it's working since I don't test it myself, but IIRC it's still far ahead compared to AMD especially if you look further back (as you mentioned). This is probably what linner was meaning to say? I had the impression that things were completely broken in linner's experience, due to the way the comment was phrased.

[moilami]

Does Intel really make CPUs?

[Paradigm Shifter]

A fake raid is lost when you switch your motherboard

Aware. Basically it was used as a cheaty way of getting large datasets into one place locally. Didn't worry about losing the array if replacing the motherboard as had backups on a large NAS, but the network wasn't fast enough to support mounting the NAS and accessing via NFS.

We've since had a network upgrade.

In my experience mdadm RAID is way more solid than any motherboard RAID and even some RAID cards. "mdadm RAID but the array falls apart on every reboot" this sounds more like you have not set it up properly or the on board SATA controller is faulty. You can also look at ZFS as a really solid alternative but it requires some additional installation under Linux.

Heh. No "thumbs up" smiley...

If you are talking about fake RAID (i.e. the RAID that is typically enabled in BIOS) then just don't use this RAID. It doesn't matter if its AMD or Intel, fake RAID historically has been trash for various reasons. Either get a HBA or in your case (since you don't have room) then just use software raid, mdadm or solutions like BTRFS/ZFS.

Ditto.

That's what I do.

If that's not a ZFS success story I don't know what is.

And again.

Thanks folks. Did some tinkering with ZFS to learn how it behaves. Initially had a go using a bunch of small fake disks in VirtualBox then dug out an old box which isn't up to much right now, stuffed a few extra disks in it and had a play on real hardware.

Worked a treat. Easy to control. Don't need it on the OS drive, so actual messing around during install is minimal.

As time permits I'll migrate things over to ZFS.

[mdedetrich]

Aware. Basically it was used as a cheaty way of getting large datasets into one place locally. Didn't worry about losing the array if replacing the motherboard as had backups on a large NAS, but the network wasn't fast enough to support mounting the NAS and accessing via NFS.

We've since had a network upgrade.


Heh. No "thumbs up" smiley...


Ditto.


And again.

Thanks folks. Did some tinkering with ZFS to learn how it behaves. Initially had a go using a bunch of small fake disks in VirtualBox then dug out an old box which isn't up to much right now, stuffed a few extra disks in it and had a play on real hardware.

Worked a treat. Easy to control. Don't need it on the OS drive, so actual messing around during install is minimal.

As time permits I'll migrate things over to ZFS.

I highly recommend ZFS, using it to handle around a petabyte storage and it was able to recover from a unexpected hard shut-down when multiple concurrent writes over 10gbe was happening.

The filesystem is eons ahead of anything else especially considering when it was made, it's only real weakness it's not that flexible when expanding existing pools so plan out your hard drives before

[Paradigm Shifter]

I highly recommend ZFS, using it to handle around a petabyte storage and it was able to recover from a unexpected hard shut-down when multiple concurrent writes over 10gbe was happening.

The filesystem is eons ahead of anything else especially considering when it was made, it's only real weakness it's not that flexible when expanding existing pools so plan out your hard drives before

What filesystems are good at that, though?

[mdedetrich]

What filesystems are good at that, though?

BTRFS and whatever unraid uses behind the scenes

[gigi]

all i care is whether more than 90% issues are resolved for the existing code? merely recognizing device is not enough... i think this applies to latter than former?