Basic doubt regarding Mirroring with LE based striping

[Nikhil_1]

Hi,

I have some confusion regarding Mirroring with LE based striping
(is it RAID1+0 or RAID0+1)?

Consider following case:

PVG0 -> PV0 PV1
PVG1 -> PV2 PV3

LV0 -> Single mirrored
Uses PVG0 and PVG1 for its mirror copies.

LEs in LV0: LE0 LE1 LE2 LE3

Allocation policy: PVG-strict/distributed

LE0 -> PV0.PE0
PV2.PE0

LE1 -> PV1.PE0
PV3.PE0

LE2 -> PV0.PE1
PV2.PE1

LE3 -> PV1.PE1
PV3.PE1

Can I explain this as RAID1+0 configuration?
Data is striped and each column in the strip is mirrored. So in
the above case, data is striped across PV0 and PV1 (num of stripe
colums = 0, stripe width = PE size). And for each PE in PV0 there's
a mirror copy in PV2, for each PE in PV1, there's a mirror copy in PV3.

Or can I explain this as RAID0+1 configuration?
Data is mirrored and each mirror copy is striped. So in the aboce case.
the data is mirrored accross PVG0 and PVG1, and then within PVG1,
data is striped accross PV0 and PV1 & within PVG2, data is striped across
PV2 and PV3.

Which one of the above explaination is correct? Why?
Or have I got the RAID concepts totally wrong?
If I say RAIDX+Y, does this mean RAID level X is applied first,
and then RAID level Y is applied on top of it?

With regards,
Nikhil

[harry d brown jr]

refers to a combination of mirroring (RAID 1) and striping (RAID 0). It combines the high fault protection of disk mirroring (where every bit of data is written to two separate disks) with the speed of disk striping (where a file is split across multiple disks to improve performance). It is popular for applications where high reliability and high performance are needed. However, it is expensive because it uses so many disks (a minimum of four) and its scalability is limited.

Damn, I lost the link, sorry. performance problems on pc causing me to shut things down. Don't you love m$



live free or die
harry

[Nikhil_1]

So is Mirroring with LE based striping RAID1+0 or RAID0+1 or neither?

With regards,
Nikhil

[Sridhar Bhaskarla · Accepted solution]

Hi Nikhil,

If you want to call it RAID1+0, there is no problem. Both are done together and nothing is top on another.

And the thumb rule for the performance is it depends. If your application does a lot of reading and little writing, then you will gain a lot from the mirroring. Reverse is expensive as you have to write two copies.

Extent Level striping does not really help much except that there is a less chance for you to develop hot disks. The stripe size is equal to the PE size here, so the effect of striping diminishes as the PE size increases.

-Sri

[S.K. Chan]

In the scenario that you gave (from the way the LEs are distributed) .. let me redraw the diagram .. (correct me if I am wrong) ..

PVG0-------------PVG1

PV0.pe0----------PV2.pe0 ==> set1
PV0.pe1----------PV2.pe1 ==> set2

PV1.pe0----------PV3.pe0 ==> set3
PV1.pe1----------PV3.pe1 ==> set4

it does look like RAID1+0 to me because you got 2 or more PVs that are mirrored first and then striped.

[Ashwani Kashyap]

Nikhil ,

While you have explained beautifully , I would not call it any type of RAID .

RAID is a redundant array of disks controlled by a RAID controller or some h/w and/or s/w mechanism within the array itself , independent of the OS .

While in your case , it is just plain software mirroring from the OS , with built in allocation and policies , though it still has redundant disks and controllers. But it still is a software overhead on the OS .