With virtualization practices trending toward 100%, I/O workloads are becoming harder than ever to predict and manage. Many storage providers and managers are scrambling to keep up with these highly random workloads and finding that traditional spinning disk storage simply cannot keep up. While flash based storage still comes in at a premium over its traditional counterpart, many industry factors are lowing the bar on flash. Combine this with big companies wanting to do more with big data and the All Flash Arrays (AFA) market is starting to look a lot more attractive to SAN managers everywhere.
AFAs are not simply your traditional spinning disk chassis replaced with SSD drives. AFAs are architected from the ground up to take advantage of the SSDs ultra high I/O for random read and write workloads. While being proficient at the task of handling these workloads hundreds of times faster than spinning disks, the engineers behind such technology also have to worry about the lifespan on flash chips and how many times the memory can be rewritten without degradation or failure.
Having laid the foundation for AFA vs. Disk, we can now delve a little deeper into how to evaluate an AFA device. First some important factors to keep in mind about flash. Flash is better at random access than spinning disk due to its ability to access memory immediately, flash is a complex storage medium with its own CPU, flash is faster at reading than writing data, and flash will eventually wear out. Several IO benchmarking tools can be used to reliably test AFAs, but configuring them properly to test against a proper workload is more important than the tools we choose. Worth mentioning are IOMeter for its friendly GUI and VDBench for its ability to select deduplication and compression ratios while testing.