RAID Link Blast

scottalanmiller

From SMB IT Journal:
The History of Array Splitting
One Big RAID 10: A New Standard in Server Storage
Choosing RAID in 2013
Choosing a RAID Level by Drive Count
Understanding Hardware and Software RAID
Nearly As Good is Not Better
Hot Spare or a Hot Mess
When No Redundancy Is More Reliable
Spotlight on SMB Storage
DAC: Dreaded Array Confusion
Comparing RAID 10 and RAID 01
What is RAID 100
The Software RAID Inflection Point

From MangoLassi:
Understanding RAID 5 with Solid State Drives
Why We Do No Preemptively Replace Hard Drives in a RAID Array
The Software RAID Inflection Point
Understanding Hybrid RAIDs
Matching Drives for RAID
What is RAID 10

From Other Sources:
ZDNet: Why RAID 6 Stops Working in 2019
ZDNet: Why RAID 5 Stops Working in 2009
ZDNet: Sorry About Your Broken RAID 5
ZDNet: Has RAID 5 Really Stopped Working
ZDNet: Don't Use RAID 5 on Small Arrays
Everything You Know About Disks Is Wrong
ACM: Triple-Parity RAID and Beyond
Understanding the RAID Penalty
Practical RAID Performance
Practical RAID Decision Making
Why Non-Uniform URE Distribution May Make Parity RAID Riskier Than Previously Thought
What You Need to Know About Growing a RAID Array
NetApp Implementation of Double Parity RAID: RAID-DP
Scott Alan Miller on One Big RAID 10 at SpiceCorps DFW on YouTube

Academic:
The Mathematics of RAID 6, Anvin

Discussions:
What Makes Parity RAID Safe on SSDs but Not Traditional HDs

This is the famous "RAID Link Blast" that gets used so often. I will attempt to update it when new articles are found to be added to the list. Very handy RAID reference to have.

Bud

Thanks! I was just thinking about this the other day and I need to pass this information on to a coworker or two...

scottalanmiller

@Bud said:

Thanks! I was just thinking about this the other day and I need to pass this information on to a coworker or two...

Good to have it all in one place. Probably should make one for SAN too.

RoguePacket

Is there a RAID failure rate calculator in those links? Been bugging me for a while. Could be a topic of an article to write (& expect hot buttons to be pushed on such).

Quick (& unvetted) search results—

ZFS tidbit, http://zevo.getgreenbytes.com/forum/viewtopic.php?f=4&t=1912

scottalanmiller

@RoguePacket said:

Is there a RAID failure rate calculator in those links? Been bugging me for a while. Could be a topic of an article to write (& expect hot buttons to be pushed on such).

No, no calculator like that exists in the industry. I'm hoping to be producing one for my master's thesis. Part of the problem is that some things are easy to calculate, like the URE risks of RAID 5 as this is essentially a function of the RAID mechanism. But the same risk on RAID 6 has many influencing factors so it because a very complex equation. So making a simple calculator or comparison is very difficult.

scottalanmiller

Most (all?) RAID calculators completely ignore the most significant risks (UREs on parity rebuild, availability impact during rebuild, etc.) and focus only on "background noise" risks so are worse than useless, they are actively misleading. They might be useful for RAID 1 risk calculation but even then have to take on huge assumptions like hot swap instant kick in to eliminate other factors like time to source a replacement drive (which they should, but it remains complex as that is a factor.) The calculators are typically made based on 1990s risk assumptions that are essentially irrelevant today so they are very bad.

scottalanmiller

Even for pure disk failures, to really get good numbers you have to know the exact drive and have a large scale study of that drive available. You need the failure curve and to understand the risk for a given period of time. Risks skyrocket in years three through five, for example, from disk failure which is almost unheard of from six to twenty four months.

thanksajdotcom

Thanks for the links for reference @scottalanmiller!

RoguePacket

@scottalanmiller Good bits.

Considering the "storage engineer" role continues to be an emerging one, better barometers do need to exist. If not number, perhaps one with adjective associations—"are you nuts??", "hmm, not bad", "can do that if you don't mind losing everything", etc.

Kind of saying it jokingly, but there is a need.

scottalanmiller

Just added the Comparing RAID 10 and RAID 01 link to the list.

technobabble

I have read almost all of these. Do you have any IOPS per RAID stuff?

scottalanmiller

Nothing on IOPS. That'll be a good topic for a future one, thanks!

technobabble

Your welcome. I found a site a couple years ago with an IOPS calculator per RAID type.

scottalanmiller

Just added the RAID 100 link to the list.

technobabble

@technobabble said:

Your welcome. I found a site a couple years ago with an IOPS calculator per RAID type.

Here is the IOPS calculator I was talking about.

scottalanmiller

@technobabble said:

@technobabble said:

Your welcome. I found a site a couple years ago with an IOPS calculator per RAID type.

Here is the IOPS calculator I was talking about.

That's the one that I often use.

Carnival Boy

Do SMB people still have RAID arguments? I just figured everyone was RAID 10 now.

I had a quote for a new ERP system last week and in the vendor proposal under recommendations for SQL server they've written "at least 300Gb of disk space in RAID 5". I was bit WTF, haven't we moved on from specifying RAID levels?

This is an ERP vendor, although they do IT support as well. I've no idea where they came up with the 300Gb figure from either. Disks are so cheap these days and you can get 600GB SFF disks so lack of storage seems unlikely to be a problem in a typical SMB.

scottalanmiller

@Carnival-Boy said:

Do SMB people still have RAID arguments? I just figured everyone was RAID 10 now.

Every few hours. RAID 6 still plays a huge role in SMB needs too. Far under 50%, but way more than 1%. I'd guess somewhere around 15-20% but that's scientific at all. RAID 10 and RAID 6 make up the vast majority of all SMB storage recommendations, though (assuming that RAID 1 is a subset of RAID 10 and RAID 100 is a super set that we ignore.)

scottalanmiller

@Carnival-Boy said:

I had a quote for a new ERP system last week and in the vendor proposal under recommendations for SQL server they've written "at least 300Gb of disk space in RAID 5". I was bit WTF, haven't we moved on from specifying RAID levels?

Especially from specifying the worst possible RAID level. Even in the 1990s RAID 5 was a "compromise" level. It should never be specified as a requirement, only possibly a minimum. In this case, the wording is confusing. Maybe they meant RAID 5 (or better) which is okay potentially. But requiring RAID 5 period, is bad. All depends on how they meant it to read.

scottalanmiller

Added a new reference on RAID Write Penalties today.
http://theithollow.com/2012/03/understanding-raid-penalty/