How important is ECC Memory?

Nelbert

Hi guys,

Looking at potentially building a new high capacity file server (this would be it's primary function and very likely it's only ever function for it's lifespan).

With that in mind is ECC memory vital in that scenario?
Reason being a motherboard and processor that support ECC memory (higher end motherboards and Xeon processor - i5's and i7's have the feature turned off apparently but are otherwise pretty much the same as the Xeon's) would push the price for those components alone over the 1K mark or certainly very close to it.

So for a file server is it important or is it more for Med Labs and stuff?

Like I could get a file server of the capacity we need for under 2K if I didn't need the ECC and was able to use the cheaper (but still higher end) motherboard and CPU.

Anyone built a file server with and without ECC RAM?

ressem

Google employees published a paper on the occurance of memory errors (using ECC).

http://www.cs.toronto.edu/~bianca/papers/sigmetrics09.pdf

We find that DRAM error behavior in the field differs in many
key aspects from commonly held assumptions. For example, we
observe DRAM error rates that are orders of magnitude higher
than previously reported, with 25,000 to 70,000 errors per billion
device hours per Mbit and more than 8% of DIMMs affected
by errors per year. We provide strong evidence that memory
errors are dominated by hard errors, rather than soft errors, which
previous work suspects to be the dominant error mode. We find
that temperature, known to strongly impact DIMM error rates in
lab conditions, has a surprisingly small effect on error

One of the points that they find is that the majority of errors happen on the same chips. They pass quality control and memory tests but they still are more prone to faults.

The latter is a little easier to diagnose on machines that have ECC as the BIOS on many platforms keeps a count. You'll notice that low price server motherboards these days even offer RAID-1 alike mirror memory configurations.

You've priced an Opteron based motherboard/CPU?

I'd be more reluctant to buy a consumer motherboard for a file server as some of those bleeding edge consumer boards are not reliable enough to be anything other than a gaming machine.
You'd need to add a decent grade network card (or 2 for channel bonding and other possibilities), serial port and inconvenience of more separate tools required to monitor the various components.
I wouldn't be buying one with USB3 or SATA 6 GB as the chipsets are more likely to be iffy. There's likely to be integrated devices such as sound cards that are better off than on a server.

(And the Dell /HP/ IBM server cases are easier to regularly dismantle and clean if you're in a small business environment than most white box rackmount kits.)

Nelbert

I would be getting an Adaptec RAID controller for the Hard Drive related aspects of things anyway.
I was looking at a Core i5 661 and Gigabyte GA-P55-UD6 with 12GB of DDR3-8500 (which supports ECC, motherboard and processor don't however.)
These items would cost under 700, and seem like overkill in terms of performance when you factor in the RAID card. Motherboard is dual gigabit lan but I could add another dual lan card if needed in the future.

A far less powerful Xeon that supported ECC and a ECC supporting motherboard would double the comparable price for those two items.

Haven't priced any AMD/Opteron solutions but would be open to a suggestion that would work?

The case is actually the most expensive part of this potential server (not allowing for the hard drives which will cost as many as we buy.) and I'm confident it's the right one for the job and cleaning/maintenance shouldn't be a problem.

Just looking for some advice really on whether the ECC pushing costs up is really a necessity for the intended purpose of the server?

Sparks

Nelbert wrote: »

Just looking for some advice really on whether the ECC pushing costs up is really a necessity for the intended purpose of the server?

That google paper basicly pointed out that there's at least one unrecoverable error per annum on average in the servers they were running (which were COTS hardware), and several recoverable errors; so if the server's running 24/7, you'd imagine ECC would be of benefit as a result.

Nelbert

Looks like an AMD X6 1055T combined with a Gigabyte GA-790FXTA-UD5 should get me the ECC support (actually works out cheaper than the i5 solution!) then.

Thanks for the advice and apologies for my blatant lack of knowledge and confusion.

corkcomp

im not sure it would matter too much for a file server. non ecc would be fine too. ecc memory can be problematic in sql servers and other number crunching programs running on the server