7 ms·
ECC is a feel good feature, but is mostly useless. Memory either is corrupted (which can be detected by running MemTest) or it works perfectly fine without ECC.
by mee_too 10y ago
ECC is a feel good feature, but is mostly useless. Memory either is corrupted (which can be detected by running MemTest) or it works perfectly fine without ECC. I have over 50 10+ year old Itanium servers with 192GB ECC RAM each, there are 2 ECC errors logged total, while running 24/7.
- amaranth 10y agoNot that it's likely to matter for these systems but there has been some recent work on using ECC to recover from errors caused by refreshing less often. This would use less power so I suspect we'll see someone try it in phones or laptops soon. http://thememoryguy.com/using-ecc-to-reduce-power/ http://thememoryguy.com/using-ecc-to-reduce-power/
- sounds 10y agoNewer memory uses smaller transitors which are more easily corrupted.
- resoluteteeth 10y ago> Memory either is corrupted (which can be detected by running MemTest) or it works perfectly fine without ECC. While there is reason to think soft errors are much less common than often claimed and therefore ECC memory has been much less necessary than people think (at least within the atmosphere), there's now a pretty good reason to prefer ECC memory because of rowhammer.
- msbarnett 10y agoECC has been shown to be ineffective against all but the most primitive forms of rowhammer attacks
- mjevans 10y agoIf you value the integrity of your data, particularly in longer term archive / recall / processing scenarios, then you value having ECC RAM as /part/ of a solution for detecting and correcting errors.
- Retric 10y agoThat's well below median of around 1+/year per server, and might indicate you are not logging all ECC errors found or just not using most of your memory heavily. However, ECC is much more useful for early detention of memory problems than actual recovery from memory problems. Further, if you get do large scale measurements you will get vastly higher numbers as the average is dominated by the worst offenders. PS: Radiation also makes a huge difference.
- dboreham 10y agoI run a few more modern servers and we see corrected ECC errors fairly often, but typically not randomly distributed across the boxes. This makes me suspect bad memory devices. We've replaced modules here and there with positive results. Anyway as a former memory subsystem designer and a former employee of a memory chip maker, I like ECC. It provides some protection against bus noise errors even if you have full confidence in the memory chips themselves.
- photon-torpedo 10y agoTo word your statement differently: without ECC, your memory either works fine or is corrupted. But you don't know which one it is, until you run memtest. How often do you do that? Personal anecdote: Upgraded RAM in my PC, ran memtest, everything fine. A year later, I need to move and so I consolidate all my data on a big hard disk. As the disk is new, I run some verification and notice some non-matching checksums. Long story short, the new RAM had become defective, and a new memtest showed a handful of errors after a few hours. Had I had ECC on that machine, I probably would have seen notifications about memory errors even before that day (plus the errors would likely have been corrected). If Ryzen allows us to build a powerful PC with ECC for little more money than a non-ECC one, I'll happily pay for this "feel good feature". The current ECC-capable alternative from the Intel universe (Xeon CPU & motherboard) is simply too expensive for most people.
- lewisl9029 10y ago> To word your statement differently: without ECC, your memory either works fine or is corrupted. But you don't know which one it is, until you run memtest. How often do you do that? Definitely this. The extra safety of mind offered by ECC RAM is totally worth the tiny price difference for me. The only thing that stopped me from actually using it was the lack of (affordable) CPU support. Ryzen changes that. On that note, I hope it's not long until we can easily buy ultraportable laptops with ECC RAM without paying through the nose for Intel's ridiculous prices for their mobile Xeon CPUs.
- sundvor 10y agoI can't wait to see more than 4 cores in a laptop!
- type0 10y agoWhat many don't realize is that ECC is a bit like paying for insurance just built in to hardware, most likely you won't need it, but if you can run ECC sticks other than cost there is no reason not to.
- strmpnk 10y agoECC is great for detecting errors more than correcting them. Then one can do thorough memtests and decide if it's time to replace memory modules. It doesn't really prevent problems (there are unrepairable ECC errors too) as much as it leads to early detection and repair. Well worth it when you're optimizing for MTTR. It's not a silver bullet if the user is just paranoid about data integrity (nothing beats end-to-end integrity checks built into your system, while ECC only addresses one step in the process).
- agapon 10y agoPersonal anecdote. I have a Phenom + Asus mobo system with ECC memory in it that has been working fine for years without reporting a single ECC error. And then they started. At first there was a correctable error once every few days, then few errors per day, and then almost a constant stream of errors. Fortunately, they were all correctable. From the reported ECC syndomes I deduced that it was always the same single bit lane. That's why only a single ECC symbol was affected and the errors were correctable. I reseated the CPU and DIMMs and the errors went away, and the system keeps chugging on. Not sure what it was, maybe some dust particle getting in a wrong place, maybe some oxidation. Anyway, I can only imagine how much trouble those errors could do if they were not detected and corrected. Especially, at the stage when there was a moderate number of them. I guess when they would become too many, then the system would just crash all the time.
- equalunique 10y agoInteresting! I had a similar setup using ECC memory on a GA-990FXA-UD3 with FX-8320. Unfortunately, what I discovered was, even though it operated OK, the actual ECC features weren't being utilized. What CPU / Motherboard combination did you use? I think it's exciting that your combination actually used the ECC functionality.
- sliken 10y agoHow much hardware have you used? I regularly find machine that crash, see errors reported in mcelog, and when I run memtest no errors. I'm guessing that memtest doesn't use a challenging access patterns, or something about powersaving/clocking that runs it a bit slower. In any case I LOVE ECC. Sure random radiation caused bit flips are relatively rare. But it's REALLY nice to see high levels of ECC errors when something died and not have to wonder if it's the flash drive, power supply, graphics card etc. Sure sometimes its the dimm, sometimes it's the motherboard, sometimes it's the CPU. But with ECC errors you can much more quickly/easily tell if it changes when you swap things around.