report on RAID/ZFS - need some help

R

ripken204

[H]ard|Gawd
Joined
Aug 24, 2007
Messages
1,427
I am doing a report/presentation on RAID and might get into ZFS if I have time.
I will have a few questions here over the next week.

QUESTION 2
does anyone have results of how well raid0/5/6 or raid-z/z2 scale in terms of read/write speeds as you add drives?

QUESTION 1
Currently I am trying to calculate the probability of getting an unrecoverable read error (URE) during a raid rebuild.
I think I did it properly for RAID5, but I am not sure how to go about doing RAID6.

here are a couple of my calculations for RAID5:
512Byte sectors = 4096 bits
URE = 1x10^14 bits
Sector error rate = 1/(URE/sectorSize) = 1/(1x10^12/4096) = 4.096x10^-11

7 x 2 TB RAID5
#Sectors after 1 disk fail = #disks x diskSize = 6 x 2x10^14 / 512 = 2.343x10^10 bits
P(have 1 error) = 1 - P(have no errors) = 1 - (1-SER)^#Sectors = 1 - (1-4.096x10^-11)^2.343x10^10 = 1 - 0.3829 = 0.6171

8x2 TB RAID5
#Sectors after 1 disk fail = 7 x 2x10^14 / 512 = 2.734x10^10 bits
P(have 1 error) = 1 - P(have no errors) = 1 - (1-SER)^#Sectors = 1 - (1-4.096x10^-11) = 1 – 0.3263 = 0.6737
Click to expand...

I would think that the probability of 2 drives failing for a 8x2TB RAID6 array would be:
P(have 2 errors) = 1 - P(have no errors) - P(have 1 error for N-1)
therfore:
P(have 2 errors) = 1 - 0.3263 - 0.6171 = 0.0566
Click to expand...

is this correct for my RAID6 calculation?
and I am assuming that RAID-Z would be calculated the same way that RAID5 is?
 
Last edited:
With RAID6, you assume that the two drive failures are independent events. You care about the probability of drive 1 AND drive 2 failing.

P(2 Drive Failures) = P(1 Drive Failure) * P(1 Drive Failure)

Although if you wanted to be more elaborate, you could think about them as non-independent events, though I don't know if you could find any data on conditional probablility.

1 - P(no errors) = P(error)

However the computation for P(no errors) is rather complicated.
 
i hate probability, grr...
i did find that article last night, but it's only for 20xTB.

as i look at my RAID6 calculations again, i can tell they are definitely wrong
basically i am wanting to graph a range of disk sizes, URE, #disks, and raid5 vs raid6
i might have to just say to assume the probability of 2 UREs with RAID6 is close to 0 for N drives... but I would rather have proof than assumption. time to relearn my conditional probabilities...
 
does anyone have results of how well raid0/5/6 or raid-z/z2 scale in terms of read/write speeds as you add drives?
 
ripken204 said:
i hate probability, grr...
i did find that article last night, but it's only for 20xTB.

as i look at my RAID6 calculations again, i can tell they are definitely wrong
basically i am wanting to graph a range of disk sizes, URE, #disks, and raid5 vs raid6
i might have to just say to assume the probability of 2 UREs with RAID6 is close to 0 for N drives... but I would rather have proof than assumption. time to relearn my conditional probabilities...
Click to expand...

See http://blogs.sun.com/relling/entry/zfs_raid_recommendations_space_performance1; it may be helpful.

Read/write speeds are pretty strongly dependent on application. I can't think of any good examples off the top of my head.
 
Someone on [H] seems to have done this for a particular hardware configuration: link. He gets up to about 700 MB/s read/500 MB/s write with 13 2TB Hitachi disks, and then it's more or less flat. He's using an HP SAS expander and an LSI 1068-based controller, with all the disks connected through a single wide port.
 
You must log in or register to reply here.
Back
Top