Lost the Assumption Game

[92dT96]

So I bought a FreeNAS Mini -- the one before the latest edition. ECC ram seems to be recommended quite a lot around here if you care for your data. I had assumed, at the time, that the FreeNAS Mini would have came with ECC ram because of who was selling it.. but you know how the saying about assuming goes.

A couple questions:

• Does the last iteration of the FreeNAS Mini support ECC ram? Can I simply buy the proper sticks and install them or do I have to buy the latest FreeNAS Mini (Diskless) to achieve ECC?

• If I must buy the latest and greatest: after I order and receive the new FreeNAS Mini can I simply take the HDDs from the old Mini and put them in the same order into the new one? Or is it more complicated than that? I'll forgo any bit flips that may have occurred -- my own stupidity and all..

Tiny-Text Disclaimer: The only blame I lay for this incident is on myself for not asking the right questions or double checking what I was getting

Thanks in advance to those who answer! :)

 

[DJ9]

Yes, the latest version of the FreeNAS mini uses ECC ram, and yes, putting in your old hard drives into the new system should work just fine.

iX should be able to talk you though the entire process, since your purchasing it from them. (that's part of the buying experience) ;)

Dear iX, I like swag. lol

 

[jpaetzel]

I have run FreeNAS at home and at work for a long time. My home FreeNAS box is the older mini with non ECC RAM. My work machines have ECC RAM in them now, but that hasn't always been the case.

Most of the FreeNAS devs use FreeNAS at home. Only one of them uses hardware with ECC RAM.

That said, why do we recommend ECC RAM? Why have we released a mini with ECC RAM?

It's a better choice when data integrity is your soul concern. Our goal with the new mini was to make the best FreeNAS box possible. It's fairly expensive, a Volvo is more expensive than a Kia too, but there's unequivocal data that it's safer too.

ZFS checksums your data, protecting contents on disk from bit rot. However the "right" type of in core corruption can cause problems for ZFS. If the pool space map gets hit with the right type of corruption it can catastrophically destroy your pool.

We don't know exactly what the odds of that are. They are very low. There are a lot of non ECC ZFS boxes out there. We know of a few dozen at most cases of catastrophic pool loss. Of those not all are related to memory corruption.

ECC RAM means that very low percentage turns in to a zero percentage for this one type of problem. It's not a substitute for offsite backups of your data.

 

[sef]

Only one of them uses hardware with ECC RAM.

Two, if that's not me...

That said, why do we recommend ECC RAM? Why have we released a mini with ECC RAM?

To quote the Wikipedia article about ECC:

The actual error rate found was several orders of magnitude higher than previous small-scale or laboratory studies, with 25,000 to 70,000 errors per billion device hours per megabit (about 2.5–7 × 10−11 error/bit·h)(i.e. about 5 single bit errors in 8 Gigabytes of RAM per hour using the top-end error rate), and more than 8% of DIMM memory modules affected by errors per year

Now, most of the bit flips that happen will be transient, and in an unused section of memory.

 

[cyberjock]

Now, most of the bit flips that happen will be transient, and in an unused section of memory.

What? Unused section of memory!? My system has 739MB free out of 22GB. You sure you want to make that argument? Statistically I've got about a 97% chance that a RAM error will occur in used RAM where data is residing...

 

[sef]

You sure you want to make that argument?

It's not an argument, it's a statement of fact. And yes, I am sure of it, and the statistics quoted by the OS are something completely different.

The likelihood of getting a correctable error depends on the type of memory and the operating environment (which includes things like usage, temperature, physical location, solar weather, and a bunch of other things). Most of these errors will happen while the memory location in question happens to be in CPU cache, and will be silently ignored as it will be over-written when the cache empties out.

For the rest, ECC helps. But with 22gbytes of RAM, doing uncached reads 24/7, you are likely to less than 5 memory errors a month.

I'm only commenting because someone might read your comment and think you're correct.

 

[cyberjock]

As someone who has done experiments with RAM in high radiation environments, you couldn't be more wrong. But, I'm not interested in arguing with you over it. It doesn't really matter. My concern with the whole ECC vs non-ECC discussion centers not on the occasional bitflip(which is hard to validate with any kind of certainty for any given system), but the fact that when the RAM fails you have widespread corruption that not only corrupts your pool, but corrupts your backups as well.

I'll let you believe whatever source you got your data from. But.. I know better.