New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
ECC vs. Non-ECC RAM or: How often did your bits flip?
Howdy folks,
when it comes to dedicated servers, I always was one of those who preferred the so-called "enterprise hardware". You know, enterprise HDs, ECC RAM, all those server-grade stuff.
Especially ECC RAM was always important to me. Losing data because of some bit that flipped out of the sudden was a nightmare. Now I wonder: How often does that even happen in real life? I mean - my personal computer does not have ECC RAM, still there is a lot of valuable data (for me) on it. Can someone knowledgeable shed a light on this for me or provide links to recent studies about this? Would be absolutely marvelous - thanks in advance!
Kind regards
-A
ECC vs. Non-ECC RAM
- My favorite type of RAM for servers is...77 votes
- ECC RAM36.36%
- Non-ECC RAM  3.90%
- I do not care. Whatever comes with the server.59.74%
Comments
http://www.zdnet.com/article/dram-error-rates-nightmare-on-dimm-street/
Interesting read and obviously a big + for ECC, thank you!
Rule of thumb. If you care about quality you use ECC.
Sure thing. I know the rule of thumb, but was looking for studies/numbers to have some beef in hand.
I had a few non-ECC stick fails with unintended consequences (maybe 1-2% of all my RAM over the last 5 years) - Lost a ZFS array this way.
But don't be fooled - i also had LOTS of ECC RAM fail (more than non-ECC), mostly with correction (parity) working, but sometimes it did not and caused, again, data corruption. If you want to be 100% safe you need to use enterprise HW and memory mirroring.
In practice, I had a server with 4 DDR2 regular DIMMs, Zeppelin the cheap type, in a noisy and often ovherheated DC for years. My newer ones with ECC DDR3 are not so stable, that one never failed, nor the VMs on it. I still have it at home, added a videocard to play games on it, it does very well for it's venerable age, Xeon 3070.
I don't think I've ever seen an example where ECC actually saved a machine from having to go down for a memory swap. Generally single bit errors, even if corrected by ECC, cause a system to go wonky.
Hum. I read about this security issue with non-ECC RAM a while ago. -> should use ECC now?
Yes, but what about data integrity?
Not for personal use. The benefits of ECC are really only apparent with servers powered on 24/7 for years on end. Most personal computers are powered off at regular intervals (e.g. every night, etc.) which effectively mitigates any risk.
If you want ECC on a personal desktop your options are going to be significantly limited (not to mention expensive). You need a motherboard that supports ECC, then a CPU that supports it (basically a Xeon). That's thousands of options down to hundreds.
All recent AMD CPUs and chipsets natively support ECC. On Intel you need an E3 or Avoton (besides E5/E7 obviously) but you can also use a (cheap) i3, chipset/MB availability is more of an issue but still not that bad, extra cost is less than 50EUR.
I just recently built a storage server (with 9 disks ) for home backup and media transcoding. Problem was I needed at least an i5 for my setup and I wanted to avoid AMD cores. This basically left me with some Xeon core (or equivalent) to pick from. I hated how I was now building around ECC RAM and not around my other requirements so I switched to using Non-ECC RAM in the end.
Well, I'm glad the choice was made based on a rigorous cost/benefit analysis.
Regarding AMD ECC, well, the problem is a lot of things need to properly designed and tested for ECC to actually be able to correct bit errors. It's a lot easier to design systems that are "placebo ECC" -- able to accept and run ECC memory sticks but unable to actually correct bit errors. To test whether a given system can actually correct single-bit errors, there are special tools that can perform a low-level write to the memory with a pattern that has a correctable bit error, and then see if it is corrected upon issuing a normal read. Unfortunately, these tools are simply not available for consumer-grade AMD boards, making it hard to tell whether you've got a cost-effective ECC setup or whether you're paying a little bit more for nothing at all.
That answer was as useless as answering "I don't know".
I've an old computer from before or around 2005 with DDR1 RAM. The original Samsung 512 MB RAM stick in it is still 100% fine and working without any issues. Usual DDR RAM without ECC or similar. No cooling kit around or it anything. The computer with it has been operating until 2011 when it was replaced with a new one. It still is working however I don't use it anymore.
Isn't that something rather good? I never had any issues with it in the terms of this subject and topic.
The problem with flipped bits in your RAM is that you may never know, why a certain file is corrupt or cannot be read. It could be a disk issue or something else. Most of the time, RAM as the issue comes to mind way to late. Therefore the "I used Non-ECC RAM for years without issues" argument is flawed as long as you are not 100% sure that there never ever was a corrupted file during that time caused by your RAM.
We're using mirrorred ECC memory system for our router (2x 3x 2GB DIMMs) at home. We have seen RAM fail more than often, but it never caused issues for the system, probably due to the quad-parity you get from using mirroring dual-parity CRC capable DIMMs (in this case all ECC DIMMs available). Enterprise hardware like HP DL (everything newer than G4 if I recall correctly) supports hot-plug for RAM, CPU (if shut down correctly inside the operating system) and obviously hard disks.
If you can afford to have a mirrored 192GB memory set instead of 384GB memory available on your server (more or less depending on your system configuration) and you are really willing to pay twice as much "just" for half of the RAM available in your server, THIS is the correct way to go.
Correct. Silent file system corruptions can be a pain in the ass. ZFS for example by now implements a fail-safe against this kind of corruption by double-validating the checksum of the filesystem and CRC table against the checksum table stored on each disk.
I really don't have a preference there is really no visible difference in DDR3 and DDR4 IMO besides quality.
You missed the point of this topic.
He's always out of the loop. At this point, I'm starting to think it's deliberate ...
In response to my decision on the poll " ECC vs. Non-ECC RAM".
A-ha, this clarified it...
then I recommend looking into what ECC memory actually is and what are the differences, starting with Wikipedia:
https://en.m.wikipedia.org/wiki/ECC_memory
Implying there is no operational difference is simply ignorant ;-).
Holy hell I need sleep my first response was off the wall lol, anyhow I think its better to use ECC memory if your looking for stability and quality I mean no since in using non ECC because who doesn't want quality hardware something they can depend on.
In the end, it all comes to shit happens when shit happens eh?
According to Google ECC is not THAT perfect either. Surely ECC is the way to go for production but still sometimes, for some projects it makes you wonder does it really worth going the extra mile?
I do lot's of bit flipping when storing ( in theory, for school) a negative int into ram. Gotta love dec to Two's Complement lol
This should have less to do with heat and more to do with radiation, EM and charged particles.
Also, someone rightfully noted the noisy MBs with circuits passing through very noisy places and being hair-thin at times.