1

We're planning to move from a 4-drive RAID6 to a SSD RAID 1, and I was analyzing smartctl data to see how many GB we've writen to the drives over the past year so I can estimate the lifetime of the SSDs based on their endurance rating.

I'm thinking that if 1GB is written to the RAID 6, then to put it simply each of the 4 drives incurs 500MB of writing (250MB for data and 250MB for parity), so I should take whatever number of GB was written to each drive for the year and multiply it by 2 to get the GB per year estimate for the SSDs in the RAID1.

Does this seem correct or am I missing something? I didn't know if RAID6 has some additional write amplification due to its algorithm or something like that.

sa289
  • 1,308
  • 2
  • 17
  • 42

2 Answers2

2

Your RAID6 description and translation to usage on a RAID1 is correct. It does not do extra writing as part of the RAID6 operation (in everything I've heard of) - everything needed would be done in the controller or software.

In general I'd have to guess you'd be safe. SSD MTBF is pretty high.

Check this out for fun - "The SSD Endurance Experiment: They're all dead"

Datarecovery.com MK
  • 158
  • 4
  • Okay, so if we had 100TB written over a year on this RAID 6, that'd translate to 200TB over the year for a RAID 1? That SSD test you posted was interesting. I found one looking at some enterprise drives at http://www.tomshardware.com/reviews/ssd-dc-s3500-review-6gbps,3529-4.html though it looks like its focus was Media Wear Indicator (MWI). – sa289 Jul 17 '15 at 18:11
  • 1
    Yep. If one (and roughly all) of the drives in your 4 drive RAID6 showed 100TB written over their life, that means they would have 200TB of actual user data written counted together (spread across the 4 drives, with 100TB of XOR and 100TB of Reed Solomon as well). The equivalent of that in a RAID1 would be 200TB on each member, since that's the total user data written, and it's mirrored. I see the discussion below about writing vs overwriting on an SSD and it's valid, but you can "hand-wave" that away by talking about large amounts over time like you are. – Datarecovery.com MK Jul 17 '15 at 19:20
0

Here's a good reference for you:

https://serverfault.com/a/5165/161412

Here's more details:

http://www.storagereview.com/guide/comp_perf_raid_levels.html

In short, RAID1 has better write performance; RAID6 has better read performance.

If you're concerned with SSD lifespan, you'll need to be aware of differing types of SSDs. In most cases, when you're writing stuff to storage, you're writing them down for long term. It's not so much "how much" you've written to storage, but more of "how many times" you've written to storage in the same spot (overwriting, deleting, etc).

If you want something with high performance, go SLC. SLC SSDs are rated for 100K writes.

If you want something to hold data for long periods of time, go MLC. MLC SSDs are rated for 2K-3K writes.

These numbers are an estimate per bit, not for the whole drive itself. Depending on how you manage your array and whether or not there's a lot of deleting and overwriting, an SSD could theoretically last you 10-20 years. If you're constantly deleting and overwriting data, an SSD could last you 2-3 months.

Because of the nature of SSDs, you may want to build a custom solution with a mix of SLC (expensive), MLC (moderately cost effective) and possibly SATA (very cost effective) drives. Use SLC for high performance writing; Use MLC for high performance reading, as in caching your data here to get a speed boost. Use SATA for long term storage of data; because most data that you keep on your system is going to just sit here for a long time anyway, and it's cheaper to replace a SATA drive than it is to replace an SLC or MLC drive.

If you end up going a route of mixing and matching drives, you may as well invest in a Nimble storage array; they do all that natively, with IOPS rating from 30K on their lowest model to over 100K on their performance model.

Community
  • 1
CIA
  • 1,606
  • 2
  • 13
  • 30
  • Thanks for the detailed write-up. My question's focused on translating the amount of writes from one RAID type to another, but I could see the information you posted being helpful to others who are looking for an answer to other questions. – sa289 Jul 17 '15 at 18:17
  • The data you provided in your question only went over how much data your system had written, but didn't include how much was being over-written. That's the key indicator for determining lifespan of an SSD. As for how the data gets striped, yes, your original assumption is mostly correct; the varying factor is the number of drives in RAID6 vs RAID1. – CIA Jul 17 '15 at 18:28
  • I'm pretty sure how much data gets overwritten on an SSD depends on the size of the SSD and total data written because they use a wear leveling algorithm such that if you overwrite an existing file, completely different sectors may be used so some NAND doesn't wear out sooner than others. – sa289 Jul 17 '15 at 18:36
  • I agree; the internal algorithms are designed to use least used bits/sectors first. However, this is not based on how much data you write. This is based on how much data is overwritten, which happens when you delete or overwrite; not how much data you have. If I have a 250GB SSD, and I can only fill it with 250GB of data; trying to fill it with 300GB of data is impossible. If I had a 250GB SSD, and I filled it with 200GB of data, then deleted it, then filled it again with 200GB of data, then yes, there will be roughly 150GB of bits that will be overwritten. – CIA Jul 17 '15 at 18:45