4 ms·

> “no ecc memory” or the like ECC is about long term stability and data integrity. For a router, meh, the network protocols will deal with any flips. For a f

> “no ecc memory” or the like ECC is about long term stability and data integrity. For a router, meh, the network protocols will deal with any flips. For a file server or database it's better if those random bit flips don't happen to critical data. AMD based systems can sometimes be forced to use ECC mode even if the BOIS doesn't support it. ECC is more important in systems with very large ram footprints because there's that much larger of a memory footprint area for cosmic rays to corrupt. If you've got one-two sticks of ram, and you're not running vital business data, meh, it's not required. I really like ECC. But I'm not really willing to pay a significant premium for it. I run companion systems to production out of my house, mostly development lanes comparable to production deployments. If they're down, it's really not mission critical. I also run my home security/surveillance systems. The other significant systems are those related to my children's computer lab.

bigiain 4y ago
One argument I’ve heard, which seems valid enough, is that running a homelab setup without EEC memory is a good thing, because you then need to build stuff that can handle memory errors (or at least notice and learn from them) which you will occasionally see in your half a dozen or so homelab machines. Once your deploying production work systems “at scale” with hundreds (or thousands) of servers, you will be guaranteed to see occasional memory errors (probably even with ECC). One of my first real homelab setups, was 6 raspberry pi’s, laid out as two load balancer, two web app servers, and two database servers. The “unreliability” of pi’s running everything off SD cards was a _feature_ because it gave me lots of experience in how my (work) software stack running on our typical AWS design held up in face of failures.

bigiain 4y ago

One argument I’ve heard, which seems valid enough, is that running a homelab setup without EEC memory is a good thing, because you then need to build stuff that can handle memory errors (or at least notice and learn from them) which you will occasionally see in your half a dozen or so homelab machines. Once your deploying production work systems “at scale” with hundreds (or thousands) of servers, you will be guaranteed to see occasional memory errors (probably even with ECC). One of my first real homelab setups, was 6 raspberry pi’s, laid out as two load balancer, two web app servers, and two database servers. The “unreliability” of pi’s running everything off SD cards was a _feature_ because it gave me lots of experience in how my (work) software stack running on our typical AWS design held up in face of failures.