Mirrored SSD, anyone?

[Gandalf Corvotempesta]

Anyone using a full-ssd storage? Which raid level?
Mirrored SSDs would make the same number of writes, thus the failure of both disks in a very very short time is possible.

Any suggestions? With hardware raid cards and SAS disks I prefere RAID-6, is slow, but very secure.
With ZFS and SSD which is the best way to get maximum data safety and availability ?

 

[wblock]

Mirrored SSDs would make the same number of writes, thus the failure of both disks in a very very short time is possible.

Took me a bit, but I think you are suggesting that a RAIDZ does less writing because not all of the data is written to each drive. But that ignores redundancy. Enough data is written to a three-drive RAIDZ1 to rebuild the data from any two drives, so figure 150% of the actual data size is written, a third of that to each drive. There might be additional overhead.

But how much are you writing? Even cheap consumer SSDs have lifetimes measured in petabytes.

 

[joeschmuck]

There are a few folks here who have claimed to be running SSD storage but I don't recall off the top of my head who that is.

I also see your point that in a mirror, if the drives are identical that the odds of them failing near the same time is possible. Also @wblock brings up a good point, you need to write a lot of data before a drive would fail from excessive writes. This could be factored into the design but if you just purchased something like the Samsung 850 Pro which has a 10 year warranty, well you are covered for 10 years based on the warranty part. Also, you could setup a script to periodically check the wear level of the SSD and if it crosses a limit, it could toss you an email.

I think the drive configuration will completely depend on the SSD capacity, the capacity the user requires, and the speed the data needs to be available. These are factors in the vdev/pool configuration.

 

[Gandalf Corvotempesta]

Took me a bit, but I think you are suggesting that a RAIDZ does less writing because not all of the data is written to each drive. But that ignores redundancy. Enough data is written to a three-drive RAIDZ1 to rebuild the data from any two drives, so figure 150% of the actual data size is written, a third of that to each drive. There might be additional overhead.

But how much are you writing? Even cheap consumer SSDs have lifetimes measured in petabytes.

I'm not suggesting anything. I'm just asking.
Are the petabytes lifetime a real value or just statistics?

 

[wblock]

That link shows actual tests where SSDs were written to until they failed.

 

[Stux]

Interesting point. Could be solved by using SSDs from two different manufacturers.

 

[Gandalf Corvotempesta]

Interesting point. Could be solved by using SSDs from two different manufacturers.

This is not always possible.

 

[Stux]

This is not always possible.

Then don't borrow trouble

 

[Gandalf Corvotempesta]

I was just asking.

 

[Stux]

I was just asking.

It's an interesting hypothetical issue. But then you have another hypothetical issue on top of that ;)

Keep a spare. Replace one of the SSDs with the hot spare every few months ;)

But if you were to do a pool like this, just have enough backups and redundancy for your data loss sensitivities.

 

[Gandalf Corvotempesta]

Backups are never the solution. You could have as many backup as you want, but you can't tell customers: sorry, i've lost all of your data, i'm restoring a backup from yesterday.
Backups are needed for disaster recovery, you can't design a storage based on a "backup". You have to design a storage based on data availability and backups doesn't increase the data availability.

I can keep a spare, but this won't solve the issue. In a RAID-1 both SSDs are wrote in the same manner, if one SSD is failing due to writes, the other could (I hope no) fail in a couple of hours, bacause the same write pattern was applied to both.

 

[Stux]

Then you need to look into high availability solutions.

 

[MatthewSteinhoff]

Anyone using a full-ssd storage? Which raid level?

We have a 2TB SSD pool (mirrored stripe) that hosts 15 XenServer VMs (using about 400GB) made with four cheap, off-brand SSDs (ADATA Premier SP550). That pool massively improved our virtual server performance. Bulk data, when required, is stored to conventional hard drives then the VMs mount that data.

We're comfortable using cheap SSDs because we snapshot and replicate often, tightly monitor our systems and are okay with the half an hour of downtime it would take to fall over to the replicated VM pool while the primary is offline. We could have hosted all our VMs with a pair of mirrored 1TB Samsung 850 PROs - a more prestigious brand - for the same price but preferred greater over provisioning to reduce wear and twice the IOPS.

When SMART 169 (Remaining Lifetime Percentage) gets to 30% or we are nearing two-thirds of their rated write endurance, they'll have been in service more than five years based on current usage. At that time, I'll migrate the data to biomechanical holgraphic neural nets.

Cheers,
Matt

 

[BigDave]

biomechanical holgraphic neural nets.

Which of course will become self aware sometime late in the 21st century and enslave all the people who
refuse to read the documentation :D

 

[joeschmuck]

Then you need to look into high availability solutions.

Agreed.

Backups are never the solution. You could have as many backup as you want, but you can't tell customers: sorry, i've lost all of your data, i'm restoring a backup from yesterday.

I thought we were talking about SSD's specifically. You would need a properly designed NAS system regardless of the storage medium. Also, snapshots and replication to an offsite location are the solution to realtime backups because if there were a fire or natural disaster, something other than a media failure, well you still need to have a realtime backup or at least as close to realtime as you can get.