Infinite I/O

The latency/IOPS curve, revisited in the age of flash

Posted by Sheryl Koenigsberg on Jul 2, 2018

One of the fundamental truths about storage performance is that there is a predictable curve that relates IOPS to latency. At first, latency rises gently, sometimes even imperceptibly, as IOPS rise. At some inflection point, this changes, and suddenly every incremental unit of IOPS that a system takes on results in an exponential increase in latency.

 

latency graph 1-1

This is true no matter what kind of storage system you're using. As increasingly more of our customer base turns to hybrid and all-flash arrays, we're discovering this along with them. It's not that hybrid and all-flash arrays don't have this inflection point, it's just that the point at which latency begins to grow exponentially is after delivering more IOPS than before.

latency graph 2

Infinio provides this same kind of shift function to the latency/IOPS curve. Just like hybrid and all-flash arrays shift the function to the right (more IOPS), Infinio too increases the number of IOPS a storage system can deliver before its latency begins to grow exponentially.

latency graph 3

What's been of great value to our customers is when these two functions are used simultaneously. Using Infinio alongside an all-flash or hybrid array continues to move the curve to the right, reducing latency for sensitive time-bound activities like reports, backups, and software compilation tasks. Another way to look at it is that if your all-flash or hybrid storage is already delivering the IOPS your applications need, layering Infinio can reduce the latency any given IOPS load operates on.

 

Topics: Performance, Latency