Forked from:

 

http://lime-technology.com/forum/index.php?topic=3864.45

 

The general premise is that HDD should be retired at some point. Whilst that is obvious when is that point?

 

Things to take into account:

 

Price of replamcent drive i.e. can one drive replace 4 at a reasonable cost

Price of upgrading the parity to cater for this new large drive

Why bother if the old drives work and parity protects you.

The actual cost of a drive MUST include the cost of all the server hardware and softare divided out by the max number of drives you can support with it. This calcualted correctly can surprise you.

Sometimes it might be cheaper to just build a new array for these small drives.

Should a drive be retired after xxx hours of use?

Should it be retired after it is xx years old regardless of use?

 

There are so many factors here that a simple question becomes a very complicated answer.

 

 

I retire a drive when it is falling down.

As I always have one or two old drives laying around after an upgrade, they can be use while i wait for a new drive to arrive.

 

If needed, i could mount a raid card on the unRAID to simulate a big HD for rebuilding purpose before adding the new drive, so the array will still be protected if another disk fails. 

 

You forgot other costs, such as loss of data, cost to replace data, inconvenience of dealing with loss, and cost to buy a drive "right now" and ship overnight or go buy it locally, rather than buy it when there is a deep discount via mail order with free shipping.  I don't like a piece of hardware dictating to me what I HAVE to do today. 

 

Replacing drives based on age or monitoring of SMART parameters is a cost of insurance against many of those things.

 

You also have to consider data growth.  If you have a fairly static array size (i.e. very little or no growth) your philosophy of drive replacement is different from someone who grows 5 TB a year.

 

The parity replacement issue can be addressed with HW (i.e. on mobo) RAID and use two 1TB drives to create a 2TB parity drive.... so you don't have to upgrade both parity and data to go to a larger data drive.  (Of course, if unRAID could do it all on it's own that would be nice, but I'd like RAID 6 first.)

Very interesting question, and very interesting replies.  My situation is this:

-My data grows rather quickly, something like 100-500 GBs per month.  As soon as I get an HDTV and start caring more about HD quality, that rate will likely increase.

-My unRAID server is at or close to its capacity of 8 drives.  If I bought a drive cage and another SATA card, I could cram maybe 10 or 12 drives in there before having to replace the entire case, mobo, etc.  I don't plan on doing this kind of upgrade for at least 6 months or a year. Therefore, practically speaking, I don't have the option of adding new drives, only replacing my current drives.

-The smallest drives in my array are 500 GBs.  To me, this is just on the side of being too big to just throw away.  I'll slowly start replacing them with 1.5 TB drives as the prices drop (just ordered my first one yesterday!).  Perhaps some day I'll build another server out of the unused 500 GBs (haha, out of habit I initially wrote that as 'unUsed').

 

Given those conditions, I'm inclined to take this approach:

"Why bother if the old drives work and parity protects you."

 

I'll only retire drives if they fail or if I really need the space, and therefore need to upgrade.

 

I'm curious about the concept of calculating the entire cost of the server per GB of storage, as you mentioned.  How do you go about doing that?  I assume there is more to it than just adding up the cost of the hardware and dividing by the total storage capacity...

Given those conditions, I'm inclined to take this approach:

"Why bother if the old drives work and parity protects you."

I'll only retire drives if they fail or if I really need the space, and therefore need to upgrade.

 

I bother because of the pain and/or time it may take to deal with the failed drive or possible data loss.

Plus the fact that it costs to spin more low density spindles.

if I had an environment where the data never grew, then I might just wait for drives to fail as long as.

1. I had a spare drive on sight just in case.

2. Removable tray less sata units (no more screws for me).

3. I exercised all the drives via smart and periodic parity checks.

4. I could turn the system off or the system was very power efficient.

5. Space permits.