Need for ECC

[NASbox]

I'm wondering what the Hardware GURUs have to say about this:

https://youtu.be/vxFNBZIAClc
https://youtu.be/52x4PSxbjUg

I can also say from personal experience in the 4 years I have been running FreeNAS I likely had a couple of memory problems for about 2 of that, and never lost any data (I'm sure a bit of luck goes with that).

 

[Ericloewe]

https://youtu.be/vxFNBZIAClc

Well, that is traumatic enough that the system will be destroyed before much can happen. It's vaguely possible to end up with corruption in such a scenario, but it should be fairly limited. It doesn't really give any insight that is useful for the whole ECC discussion, because we're well beyond "two bits wrong in 72" territory.

 

[NASbox]

Well, that is traumatic enough that the system will be destroyed before much can happen. It's vaguely possible to end up with corruption in such a scenario, but it should be fairly limited. It doesn't really give any insight that is useful for the whole ECC discussion, because we're well beyond "two bits wrong in 72" territory.

Thanks for the reply... it seems that the need for ECC is overstated in a home usage scenario that the risk of power supply blowing up and smoking the whole machine might be greater than ram error?

 

[Inxsible]

Thanks for the reply... it seems that the need for ECC is overstated in a home usage scenario that the risk of power supply blowing up and smoking the whole machine might be greater than ram error?

Possibly.

But I tend to believe the manufacturers of the appliance. If you buy any device -- TV, AV receiver, refrigerator or anything else, you will try to follow the manual to the letter because you don't want to void the warranty. In this case, the FreeNAS devs recommend ECC RAM, so I'd like to use ECC RAM so that if something wrong happens, I can tell them (forums or paid support, if you have bought it) that I have followed all recommendations and still I am having issue XYZ. And they can help diagnose the issue.

If you have an issue with non ECC, most times people will tell you, try it with the recommended configuration and if the issue persists, contact us.


At the end of the day, its all about how much risk are you willing to take.

 

[wblock]

If you have good backups (not copies, backups), then it is less of an issue. You could still lose years of data after you discover corruption in current data.

Those who skip ECC often have poor or no backups. In that situation, the risk is losing all your data versus losing several hundred dollars for a motherboard plus ECC RAM that is never called to correct a bit error. Seems like cheap insurance to me.

 

[Ericloewe]

Thanks for the reply... it seems that the need for ECC is overstated in a home usage scenario that the risk of power supply blowing up and smoking the whole machine might be greater than ram error?

No, not at all.

• Exploding PSUs are a completely independent risk from those mitigated by ECC.
• Exploding PSUs are why you don't buy Super China Happy Sun Ultra Premium Quality Top Power Clean Super Stable PSUs. You buy Seasonic or Corsair or similar. Or Delta, if you want to lighten your wallet.
• The extra cost of ECC is very small, all things considered.
• Memory errors are indeed a problem, and ECC allows you to detect that problem before it causes damage.

 

[Chris Moore]

Thanks for the reply... it seems that the need for ECC is overstated in a home usage scenario that the risk of power supply blowing up and smoking the whole machine might be greater than ram error?

The kinds of errors you get from not having ECC memory are often referred to as silent errors because you don't know about them until they get to be big enough that something just doesn't work any more. That can be a crash of the computer or a image file with a line in it or an audio file that just won't play. You never know where, when or how the error might show up.
If you value your data, listen to the voice of experience. These videos you have watched are not produced by the people that make the software, they are produced by hobbyists. They have little experience and it is no skin of their nose if you lose data. They may have even lost data that they are not aware they have lost. The popper kind of memory is not that much more expensive.
Hey, if you don't want to, that is on you, just don't suggest it to anyone else. Take the responsibility for yourself, but don't push your decision on others. The guidance to use ECC is for the safety of the data and overall system reliability.

 

[ChriZ]

I personally agree with what said on the second video that scrub without ECC will not destroy your data.
I also agree with the other thing he said, which is that you should use ECC in a server.

My personal opinion is that a server is not a server if it is not made out of server parts.
In the end, you want to build a server, ok.
For something to serve safely and reliably you must get reliable parts.
A decent non server motherboard will cost about let's say $100.
A decent server motherboard will cost about double that amount.
All other parts are about the same price.
(ECC Ram is how much more expensive than non ECC? 10% ? So what?)
With the non server mobo you certainly get lesser components, you might also need to pay another 30 for a decent NIC.
The final cost difference ends up insignificant.
You need a (file in this case) server? You should build a proper one. Not for FreeNAS, not for ZFS, but for everything else which is going to serve data. That's my personal opinion.
And once again I believe that there is absolutely no chance that ZFS will destroy my data, even with no server parts. The thing I believe is that a server should be built properly.

 

[NASbox]

No, not at all.

• Exploding PSUs are a completely independent risk from those mitigated by ECC.
• Exploding PSUs are why you don't buy Super China Happy Sun Ultra Premium Quality Top Power Clean Super Stable PSUs. You buy Seasonic or Corsair or similar. Or Delta, if you want to lighten your wallet.

Absolutely no argument there. The difference between a "Super China Happy" and a Seasonic is likely less than $100. I agree that they are an independent risk, but why I memtioned it is that I suspect that the risk of a PSU failure is statistically much greater than memory issues.

No, not at all.

• The extra cost of ECC is very small, all things considered.

The problem is not the cost of the ECC memory, but the processor/motherboard to go with it. On my current build I'm looking at about $1000 - server grade HW is very hard to come by in Canada and carries a huge markup.

No, not at all. Memory errors are indeed a problem, and ECC allows you to detect that problem before it causes damage.

Agreed it's much better, but based on those videos and my own personal experience, I think its worth the risk to go without as long as you burn in well, test regularly, watch your logs and of course use quality parts - even if they are "consumer grade" parts.

 

[Stux]

I suspect that the risk of a PSU failure is statistically much greater than memory issues.

Not in my experience.

Guess it depends on how many Super Happy PSUs you use.

 

[Inxsible]

Absolutely no argument there. The difference between a "Super China Happy" and a Seasonic is likely less than $100. I agree that they are an independent risk, but why I memtioned it is that I suspect that the risk of a PSU failure is statistically much greater than memory issues.

The problem arises when that Super China Happy fails and takes down the RAM with it.

 

[Stux]

The problem arises when that Super China Happy fails and takes down the RAM with it.

Not really, it'd take down the CPU first

 

[NASbox]

If the 5v rail goes to 12v (or higher), almost everything goes pop at once!

Having agreed that cheap power supplies are a significant hazard, I'd like to move on to some solid technical discussion of the videos in the original post. No flame wars, but a good engineering discussion/debate.

Unless there is something very wrong with those videos it make value equation VERY HARD to swallow for a HOME user.

From what I read on the forum, SuperMicro is considered the best for quality, and at least from NewEGG in Canada (one of the few places to get SuperMicro), a lot of the boards have to be shipped back to the US for some sort of firmware updates adding 2-3 weeks and another $50-$60 shipping costs to an already exorbitant price tag.

As an aside, I'm wondering if there are any good used CPU/Motherboard combos that are easy to pick up used (i7 3770 performance or better and able to accept a minimum of 32GB (64GB preferred).

 

[Stux]

Empirical evidence on the forum shows that people end up with pool corruption without ECC, even though they have plenty of redundancy.

What seems to happen is the data is written out corrupted, and then generates triple checksum errors. Meaning loss of redundancy even on a z2 array.

It’s fairly common, and when it happens you can generally predict that the user didn’t have ECC

I have never personally seen evidence for the Scrub O’ Death.

Use ECC, avoid pool corruption.

 

[nightshade00013]

Use ECC, avoid pool corruption.

Yep, better to be safe than sorry. You can not account for EVERYTHING but the better you adhere to recommended specs the less possible problems you may end up with.

As far as Canada being more expensive, I don't see how that can be a super major issue since there should be little to no problem importing from the US. And if there is, or you don't want to deal with them then just use a service http://www.borderlinx.com/CA/en to deal with the issues.

 

[pschatz100]

Unless there is something very wrong with those videos it make value equation VERY HARD to swallow for a HOME user.

From what I read on the forum, SuperMicro is considered the best for quality, and at least from NewEGG in Canada (one of the few places to get SuperMicro), a lot of the boards have to be shipped back to the US for some sort of firmware updates adding 2-3 weeks and another $50-$60 shipping costs to an already exorbitant price tag.

As an aside, I'm wondering if there are any good used CPU/Motherboard combos that are easy to pick up used (i7 3770 performance or better and able to accept a minimum of 32GB (64GB preferred).

Usually, I don't bother responding to posts like this, but a couple of statements caught my eye.

Yes, SuperMicro is well regarded on the forums. It is also well supported and there are lots of folks with experience that can offer advice when problems occur. However, there are plenty of other options. Want to save some money without compromising the integrity of your build? Consider hardware that is one or two generations old.

Why argue about the necessity for ECC ram? If you understand how ZFS works, then you will know why ECC ram is recommended. Don't want to use it? OK. The world will not come to an end if you don't. However, if a problem arises with your data, however unlikely, then be prepared to deal with it.

I never understand why folks talk about the value proposition for a home system versus a business system. I have family photos and documents that I would not want to lose. To me, they are precious. And my time has value as well.

I am prepared to spend a little money for a system that is reliable and reasonably easy to maintain - which is why I went with FreeNAS in the first place. Otherwise, I would have just purchased the cheapest components and crap hard drives that I could find and used one of the other systems out there.

And don't forget to make backups!!

 

[Evertb1]

When I started to experiment with FreeNAS I did that with a cheap motherboard with an 1150 socket, 16 GB of non ECC memory and a fairly cheap i3 CPU. Nothing to break the bank (I owned it already anyway). However as soon as I decided to become serious with FreeNAS and after a lot of reading, I knew that I should spend some money to buy the right hardware. If it's worth doing it, it's worth doing it good.

I don't need empirical evidence of any kind (espicallly not with my own data) of bad things that can happen if I don't follow up on the recommandations. I love my data. Sure, long before I started with FreeNAS I had good backup habbits. But your first line of defence is keeping your data healthy in the first place (as good as you can that is). The ZFS file system is a good start. But it comes with a price: the need of adequate hardware. And I accepted that price.