Me and my friend were discussing this the other day about how he said RAID is no longer needed. He said it was due to how big SSDs have gotten and that apparently you can replace sectors within them if a problem occurs which is why having an array is not needed.
I replied with the fact that arrays allow for redundancy that create a faster uptime if there are issues and drive needs to be replaced. And depending on what you are doing, that is more valuable than just doing the new thing. Especially because RAID allows redundancy that can replicate lost data if needed depending on the configuration.
What do you all think?
I wholeheartedly agree with you. It is worth noting that a lot of the use cases of RAID can now be solved via software, but there are some places where hardware RAID still shines, such as redundancy. Yes, software also can provide redundancy, but I still haven’t seen a software solution that is equivalent to a proper RAID controller with a dedicated battery to keep the I/O buffer alive in case of hardware failure. That one has saved me a few times.
Source: I’m in charge of 6 storage clusters at work. Beegfs is what takes care of the actual clustering, resulting in each cluster clocking in at 1.2PB of storage. Each cluster consists of four machines with three storage volumes each.
Each storage volume consists of 12 drives in a RAID6 configuration.
I can yank faulty drives and toss them out and have them replaced with no downtime. I know some like to set up hot spares, but I for one don’t. I’ve even had entire servers die on me, and thanks to additional redundancy provided by beegfs, I’ve changed motherboard with no cluster downtime either. Just move the drives over to an identical machine (yes, each cluster has a dedicated spare machine), import the RAID, and you’re good to go.
Unless I’m misunderstanding, that sounds like you’re worried about the write hole, which RAIDZ doesn’t have
It’s mostly a matter of making sure any writes that are interrupted part way through (power failure, etc) are kept alive until the issue has been resolved. The raid controller caches everything until the write is complete.
It’s not so much about disks being out of sync, but more about preventing data loss.
RAIDZ is copy-on-write, and will notice and correct parity discrepancies if interrupted partway through. Doesn’t help if you don’t get at least one copy of the data written, but I’d take RAIDZ and a UPS over a hardware raid any day
And at the scale I’m operating, I’ll take hardware raid over raidz any day. I did some performance benchmarking when initially building these clusters, and beegfs really doesn’t like raidz.
I use raidz at home, though.