JohnthePilot

This thread is for posting comments on the article Information and Strategies for using RAID. To read the original article, please follow this link.

Well done Matt. An excellent article that even I could follow.

Chevy

Very nice article, but I found one error (semantics, but an error)

Mean Time Before Failure (MTBF) - you have it as MBTF ...

The MTBF is not altered in any way, regardless of the RAID type implemented. The Real Time Before Failure may change, as a drive may be under heavier average use than if used as a stand alone drive. The Mean Time is a constant - X hours of actual usage.

JohnthePilot

I've changed the MTBF, but I'll let Matt deal with your comment.

chauffeur2

Very Well Done Matt!

A very informative article too.

Kind Regards,

TheMatt

Thanks guys.

Thanks for the correction Chevy, I guess I should have been more awake at 2:00 AM.

When I meant MTBF I was referring to the failure of the entire array and therefore the loss of data. As you add more disks to an array like RAID 0, the likely hood that one disk will crash thus destroying data on the entire array increases.

Chevy

Ahh .... I see where you went ...

Just thought of this (you may want to add this at the end) - since so many folks look at dollar amount rather than value for dollar, software controlled RAID solutions may look attractive, but they are not. The negative impact on system performance is enough of a reason not to implement software RAID, but the chance of corrupting the array is increased.

Zazula

Sorry for nitpicking (with many years of a career as a QA/QC expert I couldn't do otherwise, anyhow ), but MTBF stands for Mean Time Between Failures. It is used for repairable components only, along with MTTR (Mean Time To Repair) and MTTF (Mean Time To Failure). For non-repairable components, only MTTF is used.

MTBF is the expected time between two successive failures of a system. Therefore, MTBF is a key reliability metric for systems that can be repaired or restored. MTTF is the expected time to failure of a system. Non-repairable systems can fail only once. Therefore, for a non-repairable system, MTTF is equivalent to the mean of its failure time distribution. Repairable systems can fail several times. In general, it takes more time for the first failure to occur than it does for subsequent failures to occur. Therefore, MTTF for a repairable system can represent one of two things: (1) the Mean Time To First Failure (MTTFF) or (2) the Mean UpTime (MUT) within a failure-repair cycle in a long run.

For reparable items in an environment without a repair queue, we have MTBF=MTTF+MTTR. Availability is thus A=MTTF/MTBF=MTTF/(MTTF+MTTR). If there is a wait queue for repairs, then Mean UpTime (MUT) and Mean DownTime (MDT) should be used instead.

Specialized QA/QC applications call for specialized terminology (adapted to the specific application at hand), and hence we can have MTTR meaning Mean Time To Replace (i.e. a HDD in a RAID array) rather than "To Repair", and also terms as Mean Time Between Data Loss (MTBDL), Mean Time Between System Aborts (MTBSA) or Mean Time Between Critical Failures (MTBCF).

QA = Quality Assurance
QC = Quality Control

Additional reading:
Reliability and MTBF Overview
Calculating the True Reliability of RAID Systems
RAID Reliability
RAID reliability calculations
Reliability and Security of RAID Storage Systems and D2D Archives Using SATA Disk Drives
Computing Storage Safety
Reliability Analysis

TheMatt

In every computer book I have read, MTBF refers to the average time before a drive fails (though this number is grossly overestimated and usually not important. My usage of it is simply in the context of array fault tolerance

Zazula

Unfortunately, though, the terminology used by the HDD (and PC in general) industry's marketing departments has little to do with science... The MTBF is a statistical probability calculation tool. Have a look at this paper from the UMD: It states that HDDs are moving away from the 100,000 hrs MTBF, to 1,200,000 hrs! What does this mean - that your brand-new HDD won't fail within four months? Tough luck!

You will also be extremely surprised to find out that MTBF has nothing to do with the designed Service Life; i.e. a HDD can have a tremendously high MTBF, yet a mediocre life expectancy (and under severe restrictions, too; like e.g. maximum temperature, limited start/stop cycles, etc). But for more on those, please read through the Reliability paper linked above.

Last but not least, a RAID array is a System, not a Component; and therefore its reliability is governed by specific formulae (found and analysed on the "additional reading" links I provided in my previous post).

TheMatt

Should I change it to "Array failure probability"?

TheMatt

Notes:

"MTBF" changed to "Failure Probability"
"Performance Increase" fixed to "Medium" on RAID 0+1
"MTBF Increases with each additional disk" deleted from RAID 5 as con.

Credits: Thank you to all who participated in the discussion above about the use of MTBF.

JohnthePilot

Have you changed the copy in the New Area?

TheMatt

No, should I just copy over everything from the article to the Info Center article?

JohnthePilot

You can either do that or make the same edits in the article.

Kalim

Great article Matt!

Accessibility in general horrified me until now that I've spent 5 minutes of good searching to land to this article in it's better format. I preferred it the previous way without having the second separate accessibility link.

It seems there's 2 separate versions of the same article accessible and the majorities will land on the second, which is the least user friendly, and the first, the least user accessible: one of them looks fine but on the other, the page view is shrunken in width to accommodate for Matt's broad signature banner.

I've also noticed some other formatting errors with articles in the second link, accessed following >Info>Article>. Looks quite a buggy implementation in the layouts.

Anyway, one of my suggestions that I couldn't offer on time was (since I recall you mentioning about a continuation) that you write the first article based on single layer RAID (0, 1, 5), the second article on multiple layer RAID (01, 10, 50) and the third on special non-standard level RAID (1E, 5E, 5EE), if you could. Might be something you can possibly consider.

As far as the MTBF goes, it's just a value that means very little to us consumers in predicting a drive failure. It's there for industry standards and statistics, and that's about it really. Actual variable condition experience and mass system building failure feedback goes the longest and most accurate route for us to make use of.

Antec SP2 and TP2 had very lengthy MTBF electrolytic capacitor ratings, 15 years @ 45ºC IIRC, but a little heat and they blew up inside a month. Similarly but on a different plane, humans have a MTBF rating of around 1000 years.

TheMatt

Thanks Kalim. I was hoping to keep the article simple. I am currently working on the laptop hardware article. Maybe when I am done with that I will touch up the RAID article. I will have to find out how to split the article into multiple pages with the vB articles editor.