[Techtalk] ECC Lesson :-) [Was: AlphaServer ... ECC error"]

Hamster hamster at hamsternet.org
Thu Jul 18 22:28:54 EST 2002


> I can't contribute much info, but I would like to
> understand what ECC involves, what it does,
> and if it's usually worth the extra cost, etc.

OK :)

ECC stands for either error checking code or error correcting code or possibly even a third variant on that theme.

What it means is this: 
When you write something (ie to RAM) the ECC performs some voodoo on the data that you write - the result of this voodoo is then stored. 
Then when a read request is made for that data, the ECC checks the data its about to hand out and compares it to the voodoo it stored earlier.
Now comes the neat part. If the voodoo and the data match, then the data is simply allowed to be read. If the voodoo doesnt match, then the data is re-created using the written voodoo created earlier. 

I have a vague memory too that there are implementations of ecc that will write a message to a log file if a particular unit of ram keeps generating errors. That way you know youve got a hardware failure, and not bitrot.

I've only ever seen ECC used in context with RAM, but it might be available for other things as well. Really dont know though.

So as with just about everything, the increase in reliability comes at a cost of speed (throughput) and cost in $$. 

Hope this has helped a little???

Hamster




More information about the Techtalk mailing list