4 drive config
- Thread starter kuraijay
- Start date
It really comes down to how much parity you want.
zfs-raidz1 (equals raid5) for 1 drive failure, total usage of 75% = 3tb~
zfs-raid1 for up to 2 drive failures, total usage of 50% = 2tb~
Keep in mind stress on drives while resilver is going on leads to potential death of other drives especially with ssd's. So a 1 drive failure leads to high risks in your setup, if all ssd's are the same and did run for the same time, they are lickely to die at the same time. Its good to mix hdd with ssd in that case.
zfs-raidz1 (equals raid5) for 1 drive failure, total usage of 75% = 3tb~
zfs-raid1 for up to 2 drive failures, total usage of 50% = 2tb~
Keep in mind stress on drives while resilver is going on leads to potential death of other drives especially with ssd's. So a 1 drive failure leads to high risks in your setup, if all ssd's are the same and did run for the same time, they are lickely to die at the same time. Its good to mix hdd with ssd in that case.
Last edited:
or maybe raidz2, 50% overhead, slower even than raidz1, but you can survive to ANY 2 drive failure (better chances than raid1)
also never mix hdd with ssd (as in the same raid), it's a no-no for performance (but indeed, nice idea for "disaster recovery")
you can "mix" ssd with hdd by running vms on ssd and replicate the storage or keep some backups on hdd
also never mix hdd with ssd (as in the same raid), it's a no-no for performance (but indeed, nice idea for "disaster recovery")
you can "mix" ssd with hdd by running vms on ssd and replicate the storage or keep some backups on hdd
Guys, raidz2 on 4 devices? Seriously? And what is its advantage over the R10 in this configuration? =) IMHO, zfs is dead at the moment under Linux. It reduces the performance of enterprise ssd and even NVME at times, and sometimes by tens of times. It works worse on a consumer class HDD than using a USB flash drive as a system disk.
Like I've said, it's always a tradeoff between performance, data safety and many more.
Indeed, raidz2 / raid6 on 4 devices seems maybe stupid at a glance, but what if you loose 2 disks from the same mirror in a raid10 configuration ? You lose all the data! (well, you have backup, right ? so you restore ... but maybe with downtime, or maybe you have redundancy / DR at application level).
Of course, for each project one must evaluate all the nuances. But let's don't throw the stone without at least evaluating!
Also zfs is not about speed (actually one can say it's "against" speed
). But if you have a disk bitrot on a raid1 super-mega-enterprise-mega-expensive hp raid controller, how do you recover ?
Indeed, raidz2 / raid6 on 4 devices seems maybe stupid at a glance, but what if you loose 2 disks from the same mirror in a raid10 configuration ? You lose all the data! (well, you have backup, right ? so you restore ... but maybe with downtime, or maybe you have redundancy / DR at application level).
Of course, for each project one must evaluate all the nuances. But let's don't throw the stone without at least evaluating!
Also zfs is not about speed (actually one can say it's "against" speed
). But if you have a disk bitrot on a raid1 super-mega-enterprise-mega-expensive hp raid controller, how do you recover ? I get 99% performance without any tweaks. You must be doing something wrong.
Maybe you enabled dedup ? One big mistake you can do.