Creating a memory error

[Rugg]

I remember a long time ago LTT mentioned that they were trying to work with someone to develop a test to actually measure and detect when ECC is useful, seems like they never did that.

 

So I was wondering if someone else might know if its possible to just run some loop and detect bit flips that don't get past the OS.

[AbydosOne]

25 minutes ago, Rugg said:

So I was wondering if someone else might know if its possible to just run some loop and detect bit flips that don't get past the OS.

If the ECC is working, generally if it's a 1-bit flip, it will get logged by the OS; if it's a 2+-bit flip, the system will crash.

 

25 minutes ago, Rugg said:

develop a test to actually measure and detect when ECC is useful

I might be mistaken, but I think the paid version of one of the Memtest programs actually has that ability.

[Rugg]

22 minutes ago, AbydosOne said:

If the ECC is working, generally if it's a 1-bit flip, it will get logged by the OS; if it's a 2+-bit flip, the system will crash.

And without ECC how easy should it be to cause a bit flip? I assume if I just make like a million integer array full of the number zero, they are all going to be zero, every time, or will it sometime not be zero?

[WereCatf]

4 hours ago, Rugg said:

And without ECC how easy should it be to cause a bit flip?

Pretty easy, see e.g. the Rowhammer-exploit.

[Sauron]

9 hours ago, AbydosOne said:

if it's a 2+-bit flip, the system will crash.

Just out of curiority (the wikipedia page doesn't mention this), is this because ECC ram detects the error and shuts down the system? Bit flips on their own may or may not cause a system crash

[Rugg]

After watching this video (super interesting btw, HerrSmatzeR#7054 put it in the ltt discord), it seems reasonable that if you just fill the ram up with data, its possible you can get a memory error and if you read it back you should get a different result.

 

[WereCatf]

20 minutes ago, Sauron said:

Just out of curiority (the wikipedia page doesn't mention this), is this because ECC ram detects the error and shuts down the system? Bit flips on their own may or may not cause a system crash

It's technically not the RAM itself that shuts the system down, it's the OS kernel that halts the system in case there is an uncorrectable ECC-error.

[Unimportant]

1 hour ago, Sauron said:

Just out of curiority (the wikipedia page doesn't mention this), is this because ECC ram detects the error and shuts down the system? Bit flips on their own may or may not cause a system crash

On x86 it causes a machine check exception, for which the OS can install a handler. Windows will typically display a blue screen (which is still better then unknowingly working with possibly corrupted data). But on mission critical systems the OS/application could attempt recovery.