Thanks, Chris!
On Thu Mar02'23 02:49:49PM, Chris Adams wrote:
From: Chris Adams linux@cmadams.net Date: Thu, 2 Mar 2023 14:49:49 -0600 To: users@lists.fedoraproject.org Reply-To: Community support for Fedora users users@lists.fedoraproject.org Subject: Re: software or hardware raid?
Once upon a time, Ranjan Maitra mlmaitra@gmx.com said:
Thanks to everybody. I recall discussion from several years ago on the benefits of software over hardware RAID. I had completely forgotten about UPS for this new machine. Btw, what happens if power goes out (and I do not have UPS)?
Linux software RAID keeps a bitmap of pending writes by default, which is an okay (but not perfect) mechanism to recover from unexpected shutdown. There's also an option to keep a write journal instead, but unless you put that on a separate fast device (e.g. quality SSD with long write lifetime), it'll impact performance significantly.
There are trade-offs between various types of SW and HW RAID, so really the first question would be "what are your requirements and expectations". Are you talking about a high-uptime server, or a desktop where you just want to make hardware failure less annoying? RAID (HW or SW) is NOT backups, so you shouldn't depend on it for saving your data.
Thanks, this will be a fairly high uptime machine (not allowed to call it a server here, because that is central IT's role to have and administer:-), running lots of jobs at least a large part of the time, but the RAID will be on the /. It is more to keep the machine going if one of the two / drives fail (and till such time as I can get and put in a new one).
For /home (which is where my data reside), I have 2 backups done using rsync every hour. I plan to copy the actual /home to the second one, and I was thinking that the third one would be incremental backup (kept for a year, since I occasionally realize weeks and months later that I really want a file back from long ago) or so.
HW RAID has some advantages - quality controllers will have battery-backed cache, so things like write journaling don't impact performance and recovery from unexpected power failures is basically instantaneous. For high performance requirements, there's less overhead with HW RAID (because data only has to transit the bus once, then the RAID controller has its own paths to the drives). But HW RAID typically requires odd and/or proprietary software to manage, detect failures, etc. Depending on the RAID level you are using, recovery from a failure of the controller itself can be harder too.
Btw, I still stick to ext4, largely because of inertia (and because I have used lvm in the past and hated its naming conventions, I think, but there were also other limitations that I do not now recall) and have stayed away from zfs or btrfs or lvm. I am not sure what to do now. Clearly, things have moved far on.
I'm generally in the XFS on LVM (on SW mdraid when needed) camp myself... LVM adds a significant layer of flexibility and ability, but still using more "traditional" filesystems like XFS and ext4. I had poor experiences with ZFS at a former job, and am still a little leery of some of the approach BTRFS takes.
I'm playing with adding the dm-integrity layer for my SW mdraid (so then XFS on LVM on mdraid on integrity on drive) setup as an additional check against silent drive failures, but again, unless you put that data on a separate fast SSD, it slows down performance a lot.
I see, so your recommendation is to go for xfs?
Many thanks again, and best wishes, Ranjan