In the advanced administration and troubleshooting guide netapp indeed points out that a raidgroup should not have less than three data disks. Otherwise you will have a
It used to be (afaicr) that this was due to the way the writes were queued then flushed. Doing a burst across 3 drives was a performance optimization. I imagine that if you have 2 disks then you have unbalanced writes (eg: 1 write to disk A then 2 writes to disk B). It has been a while since I heard the detailed explanation of why 3 disks, so I may be mis-recollecting :)
all "critcal" data must be on mirrored filesystems, not on raid
Hmm. I am pretty sure that mirroring falls under one of the defined RAID levels :)
http://www.acnc.com/04_01_00.html
I suppose they mean "not parity protected" rather than raid.
On filers, if you set the raidsize to 2, you effectively get a mirrored volume.
Interesting. I had not really thought of it that way, but I guess you sort of do, in a twisted way. But the reality is that you are still relying on parity protection in this scenario, even if you can survive a 1 disk failure. So you are not doing real mirroring (eg: duplication of writes & reads). The cost and performance overhead of doing this (2 drive raidgroup) solution would make me want to further investigate this mandate.
Depending on the reason mirroring is being mandated, the features in the filer can be used to counter most, if not all of these reasons.
alexei
Alexei Rodriguez wrote:
I suppose they mean "not parity protected" rather than raid.
Depending on the reason mirroring is being mandated, the features in the filer can be used to counter most, if not all of these reasons.
My boss hit me with similar requirements wanting mirroring to protect from two disk failure. If every disk has a mirror, then if two disks fail, then you always have the mirrors.
I countered by saying what if both disks in the mirror fail?
He had no answer. :-)
He let me continue with a RAID5 solution.
Regards,
Dan O'Brien, dmobrien@lcsi.net Cell: 614-783-4859 Work: 614-476-8473 Home: 740-927-2178 Pataskala, OH
Alexei.Rodriguez@cbeyond.net writes: [...]
On filers, if you set the raidsize to 2, you effectively get a mirrored volume.
Interesting. I had not really thought of it that way, but I guess you sort of do, in a twisted way. But the reality is that you are still relying on parity protection in this scenario, even if you can survive a 1 disk failure. So you are not doing real mirroring (eg: duplication of writes & reads). The cost and performance overhead of doing this (2 drive raidgroup) solution would make me want to further investigate this mandate.
Actually, with raidsize=2 the parity disc and the one data disc will have identical contents (for those parts covered by the RAID 4 setup). Writing a stripe is in fact writing the same data on both discs -- ONTAP/WAFL uses an even parity scheme, which is why building a new volume out of already-zeroed discs is fast. Of course, that's not to say that this case of effectively-mirroring is optimised particularly well. In "degraded" mode it will still distinguish "I have a parity disc and would like to reconstuct the data disk" from "I have the data disk and would like to reconstruct the parity disc" even though these operations are essentially identical in this case.
Depending on the reason mirroring is being mandated, the features in the filer can be used to counter most, if not all of these reasons.
Indeed: it's already been pointed out in this thread that if you are rich and paranoid you can mirror plexes which are themselves made out of raid(4) groups.
Chris Thompson Email: cet1@cam.ac.uk
-
Alexei Rodriguez -
Chris Thompson -
Dan OBrien