ZFS is good and works very well if you have you disk bays fully filled. Best option, no doubts, just make sure you have enough RAM. However, if you have server with gazilion ampty bays and want to fill those bays as you go. ZFS is not for you. RAID5 extension still not possible even though author crated pull request back in 2021. So if you plan to have RAID5 and add more disks as you go LVM2 is a your choice.There are also SATA enterprise SSDs and they are not as costly as the SAS counterpart.
There is a software raid, it is called ZFS
RAID5 with LVM
- Thread starter megagolgoth
- Start date
You can always build the zRAID (as with regular RAID5) with batches of disks. RAID5 is not good (performance wise) if you just add a single disk at a time. Access pattern and amplification is going up for non-standard settings. Therefore there are "good" and "bad" setups with RAID5, e.g. enterprise setups usually have 8+1 RAID5 setups. The same applies to zRAID. I have to say that I've never extended a RAID5 group itself, I always just added another group to a RAID50. The I/O is the most important feature for me, so that may be special to my enterprise needs. YMMV. The smallest setups I had are 2+1 RAID5 groups, so just add 3 new disks and you're golden. Another option is to add and remove vdevs, so that you can change the raid type in each vdev as you go, but that does not work with a single disk.
imagine scenario. you have zRAID of 5 disk, you want to extend you add another 5 disks, in total you have 12 bays on server. what to to with another 2 bays? through away? also adding 5 disks for big companies maybe not so bad. but for smaller ones, a bit of investment.
Been there, done that ... add another shelf or have two spares available.
I get that, but I never needed to add a single disk due to the I/O performance criterion. If you need simple expandability, don't go with RAID5, go with RAID1. We seldomly use it, most software I know explicitly warns about RAID5 due to its bad I/O performance. Adding a disk to a working RAID5 will degrade performance due to the added write amplification.
Tend to agree more than disagree. however my particular scenarios, which is Proxmox Backup Server does not require significant performance. meaning I have RAID5 with max 12 rusties. But when you take into account backup service adding 20TB rustie cost you 400eur, I would much prefer to add them one by one, or three depending on the case.
But yeah, if performance is in question, I would strongly advice to skip RAID5
But yeah, if performance is in question, I would strongly advice to skip RAID5
YMMV, but for any enterprise I know, sspecially backup is highly performance critital in a disaster recovery case. Imagine you have an encryption trojan and need to roll back your whole infrastructure. Having PBS is great, because it allows to have a hot restore, in which you startup the restored VM before it is restored in order to keep time-to-restore small and everything is read directly from the PBS backend. You will have a lot of stress on your storage backend.
I totally understand that for a home setup, yet companies that cheap out on backups, are no companies I want to work with (of for)
Accept that, good enough justification. I work mainly with Azure, and backup in Azure not performance oriented. anyway if business has critical infra i would focus on HA investment not a PBS performance, IMHO.YMMV, but for any enterprise I know, sspecially backup is highly performance critital in a disaster recovery case. Imagine you have an encryption trojan and need to roll back your whole infrastructure. Having PBS is great, because it allows to have a hot restore, in which you startup the restored VM before it is restored in order to keep time-to-restore small and everything is read directly from the PBS backend. You will have a lot of stress on your storage backend.
I totally understand that for a home setup, yet companies that cheap out on backups, are no companies I want to work with (of for)
Companies that cheap-out on backup at the beginning of company growth are potential client for me
because they can invest in other things. Theoretically speaking if technology allows to go small steps, why not? or you can go more expensive but hassle free approach
it's called a choice...