scrub on SSDs?

[John Doe]

I have 2 consumer SSDs as NFS share, passed back to ESXi to host VMs.

Shall I have scrubs on the SSDs?

 

[kdragon75]

What does a scrub do? Answer that and you can easily answer your own question.

 

[spotcatbug]

I'd like a clear answer to this question as well. However...

My understanding is a scrub reads all the data on the disk and compares to already-computed checksums to make sure the data still matches those checksums. SSDs are fine for reading as much as you want; it's the writing to them that shortens their lifespans.

Confirmation would be very much appreciated.

 

[John Doe]

I have the same understanding as spotcatbug for the scrub. but there are several people in this forum which were telling you should avoid scrubs on consumer SSDs because it will significantly decrease the lifespan.

I just wanted to open a thread for discussions and feedback about your experience.

@kdragon75 as you may can see, I am in this forum for quite a long time, I appreciate the open culture and the willingness to help in this community. your answer unfortunately does not add any benefit to this discussion. maybe you will contribute to it in a more constructive way?

 

[Johnnie Black]

but there are several people in this forum which were telling you should avoid scrubs on consumer SSDs because it will significantly decrease the lifespan.

Scrub is read only, or mostly read only if errors are found, writes are what decrease a flash device life.

 

[jgreco]

I have the same understanding as spotcatbug for the scrub. but there are several people in this forum which were telling you should avoid scrubs on consumer SSDs because it will significantly decrease the lifespan.

Let me know who they are and where they said this, and they will be ... "corrected." ;-) (look jgreco made another stupid ZFS joke)

An SSD should be able to endure massive read workloads during its lifetime without concern.

The issue with SSD is that a read-resulting-in-checksum-error will result in a write to correct it.

This will happen during scrubs. But it also happens during normal operations; ZFS repairs data it notices is corrupt.

The upside to the scrub is that it is proactively looking for problems.

So if you put a super-low-end discount bargain bin SSD in your FreeNAS and it is throwing bad blocks resulting in hundreds or thousands of corrections during scrubs, the answer isn't "don't do scrubs" --- the correct answer is to ditch the SSD and not buy that model again.

 

[John Doe]

So if you put a super-low-end discount bargain bin SSD in your FreeNAS and it is throwing bad blocks resulting in hundreds or thousands of corrections during scrubs, the answer isn't "don't do scrubs" --- the correct answer is to ditch the SSD and not buy that model again.

that might be the reason, why someone has said this. could be caught in an endless correction loop and die because of that.
thanks for clarification

 

[jgreco]

By the time a disk gets to the point where everything's failing, it's toast anyways, so this wouldn't be a strong argument to avoid it.

There's a good argument to have a heterogeneous pool, however. You are less likely to experience similar failure modes in a similar timeframe if your controller silicon and flash chips are different technologies.

 

[kdragon75]

I have the same understanding as spotcatbug for the scrub. but there are several people in this forum which were telling you should avoid scrubs on consumer SSDs because it will significantly decrease the lifespan.

I just wanted to open a thread for discussions and feedback about your experience.

@kdragon75 as you may can see, I am in this forum for quite a long time, I appreciate the open culture and the willingness to help in this community. your answer unfortunately does not add any benefit to this discussion. maybe you will contribute to it in a more constructive way?

I am always happy to clarify or expound on a subject but not to provide answers that are readily available.

Give a man a fish and he eats for a day. Teach a man to fish and he eats for life.

 

[rvassar]

that might be the reason, why someone has said this. could be caught in an endless correction loop and die because of that.
thanks for clarification

The tendency to think in term of cylinders and sector and data having a fixed location has to be given up when discussing flash. There are no sectors & cylinders, everything is virtual. Even if the drive reports things as sectors & cylinders, it's just faking it for the existing interface & software.

The individual memory cells wear out. But drives have a concept of "over subscription", which is extra flash to perform wear levelling with. It can be substantial. A 1Tb SSD might contain fully 2Tb of flash cells. This supports a three-card monte thing going on behind the scenes, where a free'd block gets erased in garbage collection, and then reused. This hidden layer of indirection allows the controller to map a sector "bad" and replace it with one that still works. So if your drive starts showing individual sector failures, it would be because it ran out of good memory to substitute with, and could no longer hide the errors (or the controller itself failed...).

 

[jgreco]

Give a man a fish and he eats for a day. Teach a man to fish and he eats for life.

Steal someone else's fish if you're lazy.