Steps to Optimize the Flash Storage in Your Data Center
As flash-based storage becomes a larger and larger part of the marketplace for data storage solutions, new processes that take its unique properties into account are necessary. This is especially true at the enterprise level, where efficiency and optimal system design have a large effect on the performance and utility of data center resources. When you are integrating flash storage into your data center, these key steps will help you to find the right balance, whether your entire system is going to flash or you are adding a few new servers to an existing array.
Step 1: Reassess Your Metrics
Traditionally, data center performance has been measured in IOPS, or Input/Output Operations Per Second. With older, spindle-based hard drives, achieving a good benchmark number meant having a large assortment of drives that were capable of staging data for fast retrieval. Since flash storage has no moving parts and benefits from a different kind of physical architecture, the IOPS numbers go off the chart, and you need to use different metrics to assess your success.
Instead of focusing on IOPS, focus on these other key metrics that you are already assessing:
Bandwidth
Latency
Block size
That way, you are able to get a better picture of your data throughput, allowing you to see if access is streamlined and if your array is giving you the best possible performance.
Step 2: Get Inside the Array
Administrators moving to flash need to understand that read/write cycles will run up more CPU usage under this kind of storage. Partially, this is the cost of doing business at a higher speed. Partially, it is due to the nature of flash architecture. Either way, there are a few key changes you can make as a storage administratorto greatly improve your overall system performance:
Add more CPUs to handle the workload.
Check for unaligned write I/Os, especially if you have an older system and your hardware is not likely to automatically detect settings.
Step 3: Between the Array and the Server
The last step to take is to optimize server function and to ensure you have streamlined the process. If you are having server-wide issues, these steps are most likely to help:
Rearrange your hardware, to ensure you are complying with the rules of affinity.
Read into your operating system’s multi-pathing processes to find solutions.
Utilize hypervisor and in-guest tuning to attain optimum results
Taking these steps will ensure the best possible performance out of your flash storage.