Network and Storage Protocols

## Maximum IOPS for a RAID-DP

Hi

I am looking for the calculation to find out the maximum IOPS that a Disk shelf  can provide , Raid type is RAID-DP .

1) is it just Number of disk X IOPS per disk

It should not be right ?  we have to consider write penalty for parity and also the Shelf Loop has a limitation of bandwidth (2G/4G)

2) How  Size of IO and Type of IO play their part ..

For example :   If we consider  50% Read and 50% write IOPS, how much IOPS we will get , and how its is calculated

3) How latency play its part in the calculation .....

I am searching a deatiled  calculation that consider these all matrix and provide the IOPS that a Shelf offer ...

Do netapp explains these calclulation somewhere , Any Doc will be gr8 help

Bipul

6 REPLIES 6

Also interested in same information.

It will good if someone post answer for this Question.

Hi,

There are two ways you can approach this:

1) NetApp sizers dealing with different applications / workloads (e.g. Oracle, Exchange, SQL, etc.)

2) rule of thumb sizing, based on average number of IOPS per disk / spindle

The second approach assumes all IOPS are random with the majority of reads. Typical 'rough' figures I've seen are:

- 180-200 IOPS per 15k spindle

- 60-80 IOPS per SATA spindle

(you exclude parity & hot spare drives from calculations)

regards,

Hi Bipul

I wrote a article about this subject with how 3140 and 3160 latency levels compare and also how to calculate the theoretical maximum IOps based on type and quantity of hard disk in a RAID-DP based aggregate.

Hope this helps

Darryl

I Just came across this calculation

Raw IOPS = Disk Speed IOPS * Number of disks

Functional IOPS = (Raw IOPS * Write % / RAID Penalty) + (RAW IOPS * Read %)

Note :- For Raid 6 write penalty is 6.

Hope it helps.

That's sounds good. How to find Write % & Read %?

seeing as most the tech support are clueless with performance analysis I will help the OP out.

Run this command

statistics show-periodic -object disk:raid_group -instance /node1_aggr1/plex0/rg0  -counter read_ops|write_ops -interval 1 -iterations 60

you will need to be in Priveleged Mode - Advanced.

From there you can work out your read/write % and actually work out how to fix it, DO NOT Involve netapp support - they will read off a script and ask for a perfstat then a week later they will say "buy more disk" rather than, oh you have high latency? maybe you need to re-balance the aggregate? or reallocate or add jumbo frames or anything constructive lol.