Please comment on Drive Upgrade Strategy

[b0m541]

Notice that this post is more of theoretical nature, you might not be interested.

 

I was thinking as newly available drives are always bigger than the year before, should we have plan for upgrading the array with every drive replacement?

 

If a drive needs to be replaced, e.g. due to:

• showing some defect, or
• need for more storage space

 

One could use the following strategy:

 

1. buy 1 drive of the economically biggest available size
2. if a parity drive has a defect, replace defective parity drive, else replace the smaller parity drive with the new big drive
3. if the replaced parity drive has no defect: if a data drive has defects, replace it with the bigger old parity drive, else replace the smallest data drive 

 

The general idea is to have the biggest drives as parity drives and to always upgrade the smallest data drives with the replaced parity drives.

 

This way the array always grows, but surely much slower than the technological development of drives, because you always firstly replace the parity drive(s). If one has 2 parity drives, one could also always replace them both, so the growth is faster, but of course, more expensive.

If one wanted drastically more space, one would have to replace all parity drives and at least 1 data drive, which might be a huge cost, depending the own budget.

 

The only down-side I can currently see is that the parity drives are already aged when they get re-purposed as data drives, but I am not sure if this is a serious problem.

 

I am sure this is not a novel idea, but I just had it and could not find a similar post here (but did not spend a high effort for finding). If this strategy has already been discussed or is well-known and documented with pros/cons somewhere, please provide references.

I would be happy if you share your thoughts and experience that are relevant to the described drive replacement strategy.

Most importantly, please also point out problems that you are seeing with it.

 

 

[falconexe]

So this is what I do, and I've got a seriously large array approaching 220TB as of today.

 

As soon as a data drive has an issue, I look for the best price per GB out there, and if it is at least 50% more space than my smallest data drive, I upgrade both parity drives and the bad data drive. I run dual parity with up to 30 devices, so I have to buy 3 drives in this scenario. Then once additional drives fail, I match the parity drive sizes until I hit that threshold again.

 

In the end, it has to make sense both in price and overall growth factor for your next jump in parity size.

 

My last time doing this, I went from dual 8TB drives parity drives to 14TB drives when my 6TB (smallest data at the time) failed. I then took the 2 8TB parity drives, ran new pre-clears and re-purposed them as data drives. No issues whatsoever converting parity to data. They are still going strong years later. So I expanded my overall storage capacity, expanded how many devices I had, and increased the size of parity in one jump. The 14TB drives were the best $/GB at the time.

 

I've used this method twice since 2016.

3TB --> 8TB

6TB --> 14TB

 

If budget is a concern, I know many people who sell off their older drives in bulk on places like eBay. You never know what you are going to get buying used in bulk like that, but most of the time, you end up ahead and get some great deals. I always by new personally, but if you do go the used route, BACKUP everything you can as a safety net.

 

Finally, I've had really good luck and saved a ton of money buying bulk new USB drives like the Seagate 8TB Backup Plus drives from places like Costco. Shucking drives is fairly easy with a basic skillset and there is literally no difference from buying them single or enclosed for me. You might get a Shingled Magnetic Recording (SMR is newer tech that is budget oriented) device vs. Perpendicular Magnetic Recording (PMR has been the defacto performance standard), but in my experience, I've never noticed a performance issue with SMR even with heavy I/O. 

 

Hopefully that makes sense...

Edited January 11, 2020 by falconexe