Which architecture (RAID Grour Pool) will give us more Read & Write IO benefits in Clarrion?

There are a lot of factors that affect performance beyond RAID type and Pool configuration.. But to get it started..

1.) Traditional RAID groups perform slightly better than Pools when all else is equal (RAID Type, Disk Qty, IO Load) due to the additional metadata overhead of tracking the 1GB slices of data in Pools.  And Thick Pool LUNs perform slightly better than Thin Pool LUNs due to yet another layer of metadata for tracking the 8KB thin LUN blocks inside the 1GB slices.  Traditional RAID groups do not have either of these metadata layers.

2.) RAID 10 has the best random write performance, RAID5 is slower than RAID10, RAID6 is slower than RAID5.  The difference is due to the read/write activity required to calculate parity for RAID5 and RAID6 when writing new data.

• Example, for a single write IO from a host, RAID10 commits 2 writes to disk, RAID5 performs 2 reads, calculates parity, then 2 writes (total of 4 IO's to disk) and RAID6 does 3 reads and 3 writes.  So as you move from RAID10 to RAID5 and to RAID6, the overall load on the backend disks increases, as well as CPU utilization.

3.) RAID5/6 usually produce better random read performance compared with RAID10 using the same number of disks.  That's because only half of the disks in a RAID10 actually perform the read activity, where as all but 1 disk (RAID5) or 2 disks (RAID6) contribute to reads for parity RAID types.

• Example, RAID5 9+1 uses 10 disks, 9 of which handle any given read IO.  With RAID10 5+5 (also 10 disks), only 5 disks handle any given read IO.

4.) For sequential reads, RAID5 and RAID6 will perform similarly, RAID10 falls a bit behind, again due to the lower % of disks in contributing to the performance.

5.) Sequential writes are funny..  For a true sequential write operation that consumes at least the full size of the RAID stripe size (64KB * qty of data disks in the RAID Group), you will usually get better performance from RAID5 than RAID10.  Full Stripe Write cache optimizations in RAID5 allow the array to calculate parity on the stripe in cache before it's written to disk, without having to first read any data from disk.  This means you can write a sequential stripe to disk in one IO rather than 4 for RAID5 Random writes or 2 for RAID10 writes.  This optimization works with RAID6 as well but there is more parity so performance is not quite as good as RAID5.

• Example. RAID5 4+1 has 4 data disks (stripe of 256KB).  A 256KB write IO comes in from the host, the array calculates parity in cache, then writes 320KB to disk. (125% of the host IO size).  For RAID10 4+4, there are still 4 data disks, but the array has to mirror the IO, the same 256KB host IO requires two 256KB writes to disk for a total of 512KB.  At this point you are wasting money on the extra disks required for RAID10, and increasing the backend BUS load which could limit overall array performance.

Anyway, that's a primer..  Let me know if you have any, more specific, questions.