Hacker News Reader

Comment by maccard | Hacker News Reader

I can’t soeak specifics about the PSU but to get a server class intel motherboard (Xeon support, ECC ram), you really do have to spend quite a bit more than for the equivalent consumer board. Even if you reduce the cost of the motherboard to 200, and the psu to 150, you’re still not saving a huge amount off 5k, and you’re not getting an all in one

View on HN

Now if only they also made a laptop with ECC RAM.

I'd gladly pay 1k more to have a stable system.

It's your system really that unstable?

If so, are you sure it's the RAM?

Systems without ECC do crash, have transient "weird issues" and corrupt data significantly more.

If stability matters to you, you should really demand ECC for any system you purchase. It's the best bang for buck to that end.

> ECC ram

I'm not sure how Apple has people convinced they need ECC. It's practically worthless outside of very specific workloads. A workstation user does not fill several gigabytes of RAM with terse important data, query it millions of times/day, and do so 24/7 all year long.

It only takes one bit wrong to fuck your file system.

You make it sound like filesystems are so vulnerable. They aren't.

The chances of memory failure ruining data are extremely slim; lower than for hardware failure. If you care about your data, back it up.

Backups don't do anything if the original material contains the flipped bits.

I ran a lab notebook in word for a number of years and the document got fried by a filesystem error. The backup was toasted as well. This was corruption which broke the stage machine or parser in word which acted as a damage amplification.

As for the source of the error, we can only speculate but the possibility of a flipped bit is higher than you think.

Try running memtest86 on a PC without ECC for a week. You will be surprised at the results. I’ve seen perhaps 3 or 4 failures of this in new hardware. DRAM is not perfect.

In 20+ years that bit eluded me...

dram cells used to be bigger.

Yes. Exactly.

Also SSDs scare me more. MLC disks have so many states stored in a capacitor that is trusted not to leak over the lifetime of the device. All it takes is a few electrons to go walkies and it’s game over.

As far as you know.

From what I understand (very roughly) with ZFS or BTRFS at least it takes one flipped bit in the particular block of memory which has just been successfully checksummed and 'approved' by the filesystem as valid for a write operation. Those are quite a few hoops to jump through so let's not head towards melodrama just yet.

That’s fine but most of us are running on NTFS, HFS+ or ext4 where it matters.

And then there’s the same problem with MLC SSDs which are just arrays of capacitors encoding that state as different charge potentials. As a former EE, capacitors leak regardless of implementation.

it only takes one bit flip in the FS cache, which is what your 'free' RAM will be used for, to make your day miserable. ECC (maybe) isn't worth it for consumers of content, but for creators it should be standard. not to mention rowhammer-like attacks which ECC makes harder though not impossible.

Just because there are real reasons to have ECC does not mean you are going to experience them, let alone experience headache because of them.

Software is fault tolerant. Small memory failures can be trivially tolerated.

ECC simply isn't worth the cost, especially for workstations.

> Software is fault tolerant. Small memory failures can be trivially tolerated.

Most software is not at all fault tolerant in this respect. It just crashes. Fortunately, most users are somewhat fault-tolerant.

I've had ECC RAM once, it failed and ECC caught it before it could damage any documents. ECC memory also keeps a log of how many corrections it's made. In a long running machine that gets hot there _are_ bit flips and they _are_ significant. Especially if you're spent days working on a video file and don't have the technical skills to fix it in a hex editor.

I've also had memory issues on non-ECC machines, it's a lot harder to diagnose and much more frustrating. ECC memory should be standard in modern machines IMO.

> Especially if you're spent days working on a video file

And didn't keep a backup. Most serious software will make incremental copies to avoid that problem too.

Seriously, hardware failures are just as likely.

ECC is useful, but isn't worth it for the price unless you actually can't be fault tolerant in any way.

Backups and ECC address completely different problems. Backup systems will happily store corrupted files into your backup disks. ZFS-style filesystem level checksums won’t catch many problems either, because they only check to see if what you wrote is the same as what was read back. If the data was corrupted in memory, it’ll happily pass you back corrupted data.

Reply on news.ycombinator.com