What is the point of ECC memory?

NathanE

NathanE

NathanE

Soldato
Joined
21 Oct 2002
Posts
18,022
Location
London & Singapore
I know exactly what it does and how it works etc, so don't worry it's not that sort of question :p

But what is the point of it?

Or perhaps a better question would be: "What is wrong with non-ECC memory in servers?"

Surely if you have a system with non-ECC memory and all of a sudden it BSODs due to memory corruption then that is clearly a serious hardware fault (or maybe BIOS configuration i.e. overclocking too high!)

It seems to me like ECC memory was designed for a bygone era where electronics were far more sensitive to EFI. I can UNDERSTAND why like the Mars Rovers, space probes and orbital satellites surely have ECC memory. But why a run-of-the-mill server on an Earthly data center?
 
I think what you should more be worried about is getting undetected memory errors, especially where the servers are business critical - and most servers will be.
 
For Tier-one vendors, ECC adds things like mirroring and RAID for when uptime is the driver.

Eg http://h18000.www1.hp.com/products/servers/technology/memoryprotection.html

HP Hot Plug RAID Memory will deliver unprecedented levels of availability, scalability, and fault tolerance for 24×7 applications. It allows the memory subsystem to operate continuously, even in the event of a complete memory device failure. RAID, in this case, stands for Redundant Array of Industry-standard DIMMs.
Click to expand...
 
So if the up time is not crucial ( say i dont care if it shuts down and restarts once in a blue moon) then i can just use non-ECC ram in my server?
 
It would have helped this guy http://blog.ksplice.com/2010/06/attack-of-the-cosmic-rays/ who got bit-flipped by a cosmic ray :p (Good read of him tracking it down)

I agree though, the practical probability of such events occurring where ECC can actually be helpful in the real world (where the underlying cause is something like external interference) is very small and probably not worth the additional expense.


Some people do fear the risk is significantly higher though and hence ECC is more crucial.

A system on Earth, at sea level, with 4 GB of RAM has a 96% percent chance of having a bit error in three days without ECC RAM
Click to expand...

So maybe it is worth having?
 
Last edited:
If you're running a server which is a single point of failure you're doing it wrong - or are in space.

In the real world, your server should be able to crap-out without clients noticing.
 
I've only ever seen it be of any advantage to us once. One of the Exchange boxes reported "Uncorrectable memory errors" and basically quarantined that part of the memory, allowing the server to continue.
 
billysielu said:
In the real world, your server should be able to crap-out without clients noticing.
Click to expand...

It's a lovely ideal to aim for but the reality is many companies particularly small ones simply cannot afford such luxury, why else would single box SBS installs be incredibly popular?

Back on topic I've never been in a situation where I've known that ECC memory has made a difference but I'm not about to advocate dropping it!
 
a1ex2001 said:
It's a lovely ideal to aim for but the reality is many companies particularly small ones simply cannot afford such luxury, why else would single box SBS installs be incredibly popular?
Click to expand...

Exactly, end of the day it's all about the £££, cost/benefit etc.
 
It should prevent undetected and non fatal memory errors, which could in the wrong circumstances be commercially troublesome - if you're running a serious financial system then you really really don't want a memory error screwing with your data, anything to guard against that is worth the money. It is somewhat redundant in most modern environments but it has some benefits and little extra cost so why bother with another SKU and supply line?
 
Any server dealing with sticky data (File serving, long term archive, email) should be using ECC.
Any machine (Server or workstation) dealing with sensitive transactional data should be using ECC.

Software handling sensitive data should have checking and redundancy designed into it, but we can't always peer inside the black box. Better to be safe than sorry - not to mention the marginal cost of ECC vs. faffing around with lost data.
 
Isn't the main point of ECC RAM that there's no point in stealing some from work :D

We have loads of this type of RAM at work just lying around albeit it's slower speeds then the popular normal RAM used in desktops and nobody ever goes near it. Pens and pencils on the other hand are a different story :)
 
IAmATeaf said:
Isn't the main point of ECC RAM that there's no point in stealing some from work :D

We have loads of this type of RAM at work just lying around albeit it's slower speeds then the popular normal RAM used in desktops and nobody ever goes near it. Pens and pencils on the other hand are a different story :)
Click to expand...

Borrow some for me...need 4 gig of pc-133 ecc sdram please!!!

:D
 
You must log in or register to reply here.
Back
Top Bottom