ZFS RAID size and reliability calculator 2017-05-17

[MtK]

@Bidule0hm, nice calculator.
it would be nice if you could add "number of VDEVs" into the calculator (stripes).

 

[Bidule0hm]

Thanks ;)

Hmm, why not but it's very complex to do the calculations then. I'll see if it's doable.

 

[soulburn]

@Bidule0hm, nice calculator.
it would be nice if you could add "number of VDEVs" into the calculator (stripes).

First of all, thanks Bidule0hm for this great calculator, it's clearly the best one out there. Second, add my vote for the ability to adjust the number of vdevs.

 

[Bidule0hm]

Thank you ;)

 

[Alf928]

Many thanks for this calculator, which I only found from your sig on my thread. Completely forgot to include the overhead and 80% rule into my disk space calculations so having this calculator showing me all of that is awesome :)

 

[Bidule0hm]

Thanks ;)

 

[Bidule0hm]

Thanks ;)

Edit: oh god, this forum engine well, sorry for the DP, you can delete this one.

 

[ovizii]

@Bidule0hm - been looking to replace some drives and stumbled upon tons of discussions about URE 10^14 vs the enterprise class of 10^15 and 10^16 so I was wondering why your calculator doesn't take the URE in consideration or does it somehow?

 

[Bidule0hm]

Because ZFS can take care of that for a healthy pool, excepted with striped drives of course.

Then the only time you need to take the URE into account is when you have a vdev in a degraded state with no redundancy left. In theory I should use the URE but it's extremely complex. If you can provide a formula (for each vdev type) who use the URE I'm more than happy to include it :)

NB: I'll move the app to my website as soon as I have the time and I'll do a lot of changes to make it better (including the exact overheads calculations now that I finally know how to do it) ;)

 

[ovizii]

Ah, I see, thanks for clarifying!

 

[Amir Yalon]

May I suggest to display the MTTDL on a logarithmic scale?

The way it is displayed now, in scientific notation, there are two numbers to compare at the same time: the coefficient and the exponent. If, instead, a logarithmic scale is used, there would be only one number to look at. (In base 10, it would be comparable to the currently used exponent. You can even show it as 1 x 10^x h, where x can be non-integer.)

Thanks, this tool is awesome.

Yours,
Amir

 

[Bidule0hm]

Good idea, I add that to the TODO list ;)

 

[mattlach]

Hmm. Number of vdevs would be great, agreed.

But also, these calculations only look at total pool loss, right?

What about the exposure to URE's during rebuild if you have lost enough drives to lose parity?

If you lose one drive with RAIDz (or two with RAIDz2 or 3 with RAIDz3) there is no protective parity until the resilver is complete, meaning that the pool is susceptible to corruption due to URE's until resilvered.

You won't lose the entire pool, but you could wind up with a few corrupt files in it.

 

[Bidule0hm]

I still need to upgrade the probability calculus but it can get really complicated really quickly... If you have the formula you are more than welcome to share it :)

 

[mattlach]

I still need to upgrade the probability calculus but it can get really complicated really quickly... If you have the formula you are more than welcome to share it :)

Well, the URE probability is usually expressed as a probability per bit, typically 10^-14 for consumer drives and 10^-15 for enterprise drives.

So it would be a matter of calculating the probability of having a condition where drives fail such that you are down to no redundancy during resilvered, then multiplying the the total number of bits to be read during the resilver (n-1 drives)*(terabytes*1024^4*URE rate) to get the predicted number of bits corrupted.

At least I think. There is probably someone more knowledgeable about this here.

As for the details, I'm good with applied statistics, not so much with probability, unfortunately.

 

[Bidule0hm]

The main problem is to add that to the current formula... I really need to redo the app from scratch so I can add as many formula as I want without the thing to be a big mess as it would be currently :)

 

[Ericloewe]

Hey, do you mind if we move this to the Resources section? If you don't have the time to do it yourself, it's no problem, I can move it for you and set this thread as the discussion thread.

 

[Bidule0hm]

Yes, no problem ;)

Sorry for the delay, I was on vacation.

 

[Ericloewe]

No problem. Would you rather do it yourself or should I do it?

 

[Bidule0hm]

You can do it if you don't mind; I'm rather busy these days. Thanks ;)