We have written quite a bit recently about the tremendous growth of flash technology adoption across our storage system families, both all-flash and hybrid. And now that flash has been fully integrated into our core platform architectures (both hardware and software), we are spending more of our time and resources helping customers determine the best fit with their application and business requirements. For those workloads that need the lowest possible latency and highest IOPS, we recommend our EF550 all flash array but we’ve found that for most of our customers, a hybrid approach that combines spinning disk with a small amount of flash can meet their needs at a much lower price point. So, for this later hybrid approach, we are often asked how to best determine the optimal amount of flash to deploy and to ensure the best match for specific workloads.
One of the coolest new features of Data ONTAP 8.2.1 is Automated Workload Analyzer (AWA) which automates Flash Pool sizing for our FAS systems. AWA monitors live workloads in real time on a Data ONTAP system providing needed cache sizes and HDD offload percentages (hit rates for different cache sizes) as a result of the workload analysis. The result is that you can now estimate the performance gain of a hybrid array even before you deploy Flash Pool.
AWA works at the Data ONTAP aggregate level, and while it is a tool optimized for Flash Pools, it can also provide benefits for hybrid arrays based on Flash Cache. In the Flash Cache case, there are some caveats and since Flash Cache is a controller wide cache, AWA will need to be run all aggregates on the system. And AWA accounts for overwrite caching which is a Flash Pool only capability, so projected cache sizes will reflect this and will be higher when there is overwrite caching involved.
One of our key goals with AWA has been to keep things simple. With this in mind, AWA does not require any detailed understanding of workload characteristics. It can be run live on a storage system with flash installed or on one with HDD storage only. Summary output provides the following:
-Percentage of reads and writes that are cacheable
-Maximum cache size including reserve space
-Projected cache hit rates
The bottom line is that we are taking the guesswork out of deployment of hybrid arrays. The result is that you can deploy flash faster and with predictable results while meeting the requirements of your actual workload environment.