When we chart the Array’s cumulative IOPS over time, what it going on becomes clear.
Here we see 5 applications that work across the year in a cyclical fashion. Application 5 closes out each quarters books and application 4 is that quarters activity leading up to the end of the quarter. This pattern is repeated each quarter. But add to that Application 3. Here we have an IO load that is increasing with the year. Think of a new application that is increasing in popularity. This adds an exponentially increasing IO demand to the system. Luckily Application 2 is exhibiting the opposite profile. Here we see an exponentially decreasing IO demand. Very likely an older application that is finding less use as time marches on. Finally, there is Application 1. Here we see an IO demand that starts high, then goes lower. Only to increase again towards the end of the year. This makes sense: The business here is busy with Christmas returns and a January sale. During summer, activity drops. Only to resume again as the holidays approach. A good example of how data flow follows business flow. Now let’s see how the array must shift optimization focus across the year:
Note to Presenter: Click now in Slide Show mode for animation.
We start the year with the Apps all having a somewhat even demand on the storage system
Note to Presenter: Click now in Slide Show mode for animation.
But already at the end of the quarter, we now see that the new App 3 is beginning to make its presence known
Note to Presenter: Click now in Slide Show mode for animation.
And App 3 continues to have an even larger IO footprint as the year progresses
Also notice that App 1 and App 5 are dormant at this point
Note to Presenter: Click now in Slide Show mode for animation.
Finally as the year winds down we see that the IO distribution has shifted a bit. App 4 is now dormant, while App 5 is coming back and App 3 is still going strong. So what are we learning here? That Application workloads are DYNAMIC! So applying old-school static RAID-set/LUN configurations to meet IO demand is guaranteeing one thing only: that we will be WRONG!
We cannot meet the needs of a dynamic world with a static approach!!! That’s why we need FAST.