ECC RAM for server(s)

I have a few (mini)servers (1xAMD E-350 & 1xIntel i5-2500k) running 24/7 that don't support ECC RAM. However I will store data mostly on dedicated servers with AMD AM3(+) CPUs which support ECC RAM. Now the question is: since I work from my desktop on a LAN to the file server (AM3+, ECC supported - unbuffered) and all traffics pass through the AMD E-350 which does NOT support ECC, will my data be corrupted by this board before arriving on the safe ECC-supported server (traffic will pass as a linux bridge, since this will be a linux debian / gentoo router).

user51166

Posted 2011-10-12T16:45:25.273

Reputation: 538

2Uh, just because something isn't error correcting doesn't necessarily mean it's INTRODUCING ERRORS CONSTANTLY... – Shinrai – 2011-10-12T17:01:28.287

In fact I saw some comments saying that if your ram is bad then all your data could get corrupted. If it's "only" 0-10 x single-bit errors per 16GB per day (which, still, is worth to be considered) then I might be able to live with that. My primary data server (real data, not movies or anything that can be replaced) will run ecc, but all network traffic goes through a non-ECC server (which I think is the way low-cost routers work, or do they use ECC ?). The question is: the traffic gets directly bridged from one card to the other through the cpu or memory could corrupt data transmitted?Thanks – user51166 – 2011-10-12T17:08:58.090

There's a difference between "Does it introduce the potential for corruption?" and "Will my data be corrupted?". My point was that your question currently reads like the latter. – Shinrai – 2011-10-12T17:10:54.143

Well reading comments online about ECC ram seems to lead to the latter since the chance of having a memory bit error in non-ecc ram seems to be pretty high if your server is always on (even if it's one bit corrupted in 1 year it's still almost certain, although the probability of having a memory error in the next 2 hours might be very small). – user51166 – 2011-10-12T17:25:21.263

However just say what you think about it. I don't really like probabilities anyway ^_^ – user51166 – 2011-10-12T17:28:12.537

Why do you think your data is passing through the non-ECC server? – Ƭᴇcʜιᴇ007 – 2011-10-12T17:29:29.040

Because that's my router & firewall for both LAN and WAN (linux NON-ECC server) – user51166 – 2011-10-14T14:16:51.147

Answers

Your data can get corrupted (but it is unlikely to ever happen).

Linux IP forwarding operates at the Network level of the OSI model. This means that the TCP/IP datagrams (which include a 16-bit checksum field because network connections are so unreliable) will be decoded, and a new checksum will be computed for the rest of the transport. Even though the probability of an error occurring just at that time is really low, it cannot be excluded completely.

You could mitigate this risk by using a secure protocol like SSL. The data will travel encrypted and will be decoded and checked at arrival. Integrity of the data is guaranteed by SSL, but I don't know if an error will trigger a missing file or its re-downloading (probably depends on the tool you use).

Calimo

Posted 2011-10-12T16:45:25.273

Reputation: 1 119

Since this is a linux bridge, it seems that the client and server on either side of it are using a particular protocol to communicate, which doesn't care about the bridge. You need to:

use or add another protocol here that verifies data transmission
do that yourself on received data through the use of scripts or something else on client and server

So if you are transmitting files, one way to solve this is:

Client computes SHA-1 hash of all files to send
Client sends files to server
Server computes SHA-1 hash of all files it received.
Compare SHA-1 hashes afterward to verify transfer.

LawrenceC

Posted 2011-10-12T16:45:25.273

Reputation: 63 487