Benchmarking: Disk Bandwidth analysis of m instances|Part 2 FIO
In a previous post, FIO benchmark was used for four types of IO operations on storage-optimized instances:
-
Sequential Read
-
Sequential Write
-
Sequential Read-Write Mix
-
Random read-Write Mix
In calculating the geomean for each operation we made sure that the number of parallel FIO processes matched the number of available instance-vCPU cores, and that we used all available instance-store drives for measurement.
The m3.large instance provided the best FIO performance-per-dollar. Its only downside is that it offers only a single 32GB SSD drive, which may not be enough for many applications.
Our analysis showed that instances with an SSD drive offered the least latency. Average latencies for each instance type are shown in the table below. Similar to our methodology for the FIO bandwidth analysis, we also calculated the geomean of the average latencies of each IO operation.
As expected, in terms of latency, the newer instances with SSDs completely outclassed the older HDD instances.
