FreeNAS Mini review @Tom's Hardware

[Ericloewe]

Link: http://www.tomshardware.com/reviews/ixsystems-freenas-mini,4162.html

It's interesting to see just how little L2ARC is helping read throughput with just 16GB of RAM.

 

[hugovsky]

I find this comment interesting:

ammaross , July 16, 2015 7:58 AM
Quote:
utilizes the DRAM as a cache for in-flight data, you'll want to stick with ECC memory to protect that information from corruption.

Just by way of correction, ZFS has memory page checksums in addition to disk block checksums so even a memory bit corruption on non-ECC RAM can be healed on-the-fly.

 

[@ndres]

Hi.

I need the Hardware Compatibility List Hard Disks for the FreeNAS Mini device.

Thank you very much.

Best regards.

 

[DifferentStrokes]

Maybe this will help.

 

[Ericloewe]

I find this comment interesting:

It's inaccurate. The option does exist, but I'm sure it has an enormous performance impact.
And I'm not sure FreeNAS includes this option.

 

[cyberjock]

The option, in my opinion, is totally untested anyway. The reason why ECC RAM is what it is, is because software can't be trusted with things like bad RAM. Inevitably, it's possible that the precise code that is supposed to check for the corruption is, itself, corrupted. So what do you do then? How far down the rabbit hole do you go before you say "gotta handle it in hardware". The IT field has already spoken.. they deal with it in hardware. This is fundamentally true for literally 99.9% of everything that stores data.... SD cards, floppy disks, hard drives, RAM, the cache in your CPUs, etc. *everything* has some kind of checksum and repair mechanism (within reason based on the level of corruption) except for your typical system RAM.

Imagine if you want to verify there is no corruption by storing two copies of every value. So now you need "just" twice the RAM. Totally good deal right? Would you buy 16GB of RAM just so you could "only" have 8GB usable? I wouldn't. And what do you do if the two copies don't match? Do you store 3 copies of the data then?

So imagine if you have 3 copies of your data in RAM... just kidding.

I find it just a bit funny that everything else in computing has it's own way to validate information isn't corrupted, but system RAM isn't one of them. How weird/backwards is that!?

 

[Ericloewe]

Imagine if you want to verify there is no corruption by storing two copies of every value. So now you need "just" twice the RAM. Totally good deal right? Would you buy 16GB of RAM just so you could "only" have 8GB usable?

To be fair, this is a feature on Xeon E7 and Xeon E5 2xxxx (Xeon E5 only has a single unified memory controller) - the two memory controllers work in parallel, mirroring each other. And people actually buy it, because they're paranoid in the extreme. :p It's probably meant to improve uptime, though, not correct whatever ECC misses.

The reason why ECC RAM is what it is, is because software can't be trusted with things like bad RAM.

In addition, RAM is slow enough as it is! Do we really need to increase our RAM accesses by 1/8 and do simple arithmetic on that data in software? Of course not, it's slower and more expensive than just buying ECC RAM!

I find it just a bit funny that everything else in computing has it's own way to validate information isn't corrupted, but system RAM isn't one of them. How weird/backwards is that!?

Penny-pinching at its best.

 

[joeschmuck]

Not a bad review at all if you keep in mind that they were not covering all the FreeNAS software features but more just how this system runs in a basic configuration.