Full Multi-bit reliable auto-correction (RAIM) & FreeNAS

[Yuge]

Hello,

As discussed on the blog in the link below, and multiple of other places on the internet regarding large high quality server environments, RAIM and similar technology does indeed seem to be going beyond ECC memory. It is however requiring twice the amount of memory in order to create redundancy for multi-bit errors or full DIMM failures, and protection against DRAM exploits and hacks. However the gain for some uses is significantly positive for the price of double memory.
http://millennialmainframer.com/2015/03/yes-use-ibm-mainframes-reliable-secure-memory/

What I've been stuck with though, is finding any motherboard that supports this technology, it starts to seem like its only available in IBM z Mainframes? HP is talking about something called Memory Mirroring, though I have no insight in whether Memory Mirroring is worse, just as good, or even whether its better, compared to RAIM. Similar issue occur with Memory Mirroring capable motherboards, these are hard to track down as well.

1) Is this technology available for small server and workstation motherboards?
2) Does FreeNAS support RAIM or similar technology, that pushes beyond ECC memory?
3) Any advice on how to track down these motherboards? i.e. certain brands, chips or sockets?

I'm aware that ZFS+ECC memory is pretty good, though what I seek is the ability to avoid crashes, down time and processing operation looses, which is occurring outside the capability of ZFS+ECC.

Any input is greatly appreciated. If RAIM like capability turns out to be currently impossible, then at least I can stop looking, finding an answer eludes me. But it would be pretty neat if the hardware is available on the consumer market, and supported by FreeNAS.

Does anyone have some thoughts on whether RAIM or similar technology is currently possible for FreeNAS?

Happy New Year, btw!

 

[Ericloewe]

Xeon E5-2xxx and up have this sort of capability, at least on paper (or maybe it's Xeon E7 only?).

I'm aware that ZFS+ECC memory is pretty good, though what I seek is the ability to avoid crashes, down time and processing operation looses, which is occurring outside the capability of ZFS+ECC.

Mitigating the cases caused by RAM improves reliability only by a trivial amount. If you really need high availability, that's what TrueNAS is for.

 

[tvsjr]

I'm aware that ZFS+ECC memory is pretty good, though what I seek is the ability to avoid crashes, down time and processing operation looses, which is occurring outside the capability of ZFS+ECC.

I would argue that there are many, many things that are FAR more likely to cause you outages/downtime than a lack of redundant memory. Unless you're operating in an extremely high performance environment where you've already got massive redundancy from the chassis level all the way down, I'd start there.

Supermicro has supported memory mirroring (on top of ECC and SDDC) as far back as X8, with the 5500/5600 series processors. E5 supports it as well, as long as the motherboard also does. From what limited reading I've done on the technology, it's primarily aimed towards large compute clusters (supercomputers) with very long-running jobs, where a system might fail in the middle of a job and impact the entire cluster's results.

 

[Arwen]

With memory densities going to 32GB-64GB per DIMM soon, (or now?), in my opinion,
any new memory standard, (DDR5...), needs to support better than stock ECC. Specifically,
I am thinking of 80 bit wide memory being able to at least support detecting 1-3 bits of error
and correct 1-2 bits of error.

Obvously this is not as good as RAIM, or various other methods. But, it's also MUCH cheaper.
Plus, it can be combined with other methods, like bank mirroring.