Questions about LVM parity and MDADM vs. LVM as backup on another system
by road hazard from LinuxQuestions.org on (#6CKFE)
WARNING: I don't want to use ZFS so if you're going to suggest that, just keep on walking please. :)
My current setup is a main server with an MDADM RAID 6 array (XFS fs, 72TB total, 20 4TB drives) and the backup server is an MDADM RAID 5 array (XFS fs, 72TB total, 10 8TB drives). Every day, the main server backs does an rsync to the backup server.
Over the years, I've added drives to both arrays and after a day or so of non-stop re-silvering, everything just keeps humming along.
My next upgrade will be replacing all the 4TB drives on the main server with 8TB drives and just adding more 8TB drives to the backup server so I'll end up with 144TB in each chassis. I'm in no hurry as I have about 16+ TB of free space on my existing setup and based on my current storage trajectory, I'll have another year or so before I need to do any of this.
Wanting to cut down on re-silver time (due to unfounded?, scary stories about POSSIBLE read errors popping up during non-stop disk I/O of 8TB (and bigger drives), I was thinking about doing away with MDADM altogether and using LVM on both servers. If a drive fails, and brings the entire LVM down, I'll just replace the drive and copy data over and get back up and running.
BUT, if I have everything in a single LVM, will life suck trying to figure out which files are missing? Will having an LVM with parity help that?
Next question.... I wonder which is safer..... keep my existing setup (MDADM arrays on both servers) or LVM (with or without parity drive(s) on each: if a drive fails in MDADM, the hot spares will jump into rotation, re-silver kicks off and a day or 2 later, done vs...... LVMs on both and spend an hour or 2 copying data back to a failed 8TB drive and done.
Like..... in a worst case scenario..... would one solution be any safer over the other? Sure, with MDADM, I have guaranteed uptime vs LVM on both which would mean the main server would be offline for a couple of hours during the copy process if IT had a drive failure and I need to restore from the backup one.
Just looking for opinions. Something I agree with others on, increasing drive capacity makes re-silvering more and more worrying to me. The thoughts of re-silvering 2, 16TB drives sends chills down my spine so I can see why folks lean towards some other solution......... but what? :)
My current setup is a main server with an MDADM RAID 6 array (XFS fs, 72TB total, 20 4TB drives) and the backup server is an MDADM RAID 5 array (XFS fs, 72TB total, 10 8TB drives). Every day, the main server backs does an rsync to the backup server.
Over the years, I've added drives to both arrays and after a day or so of non-stop re-silvering, everything just keeps humming along.
My next upgrade will be replacing all the 4TB drives on the main server with 8TB drives and just adding more 8TB drives to the backup server so I'll end up with 144TB in each chassis. I'm in no hurry as I have about 16+ TB of free space on my existing setup and based on my current storage trajectory, I'll have another year or so before I need to do any of this.
Wanting to cut down on re-silver time (due to unfounded?, scary stories about POSSIBLE read errors popping up during non-stop disk I/O of 8TB (and bigger drives), I was thinking about doing away with MDADM altogether and using LVM on both servers. If a drive fails, and brings the entire LVM down, I'll just replace the drive and copy data over and get back up and running.
BUT, if I have everything in a single LVM, will life suck trying to figure out which files are missing? Will having an LVM with parity help that?
Next question.... I wonder which is safer..... keep my existing setup (MDADM arrays on both servers) or LVM (with or without parity drive(s) on each: if a drive fails in MDADM, the hot spares will jump into rotation, re-silver kicks off and a day or 2 later, done vs...... LVMs on both and spend an hour or 2 copying data back to a failed 8TB drive and done.
Like..... in a worst case scenario..... would one solution be any safer over the other? Sure, with MDADM, I have guaranteed uptime vs LVM on both which would mean the main server would be offline for a couple of hours during the copy process if IT had a drive failure and I need to restore from the backup one.
Just looking for opinions. Something I agree with others on, increasing drive capacity makes re-silvering more and more worrying to me. The thoughts of re-silvering 2, 16TB drives sends chills down my spine so I can see why folks lean towards some other solution......... but what? :)