Flash storage is increasingly more affordable, so much so that many organizations are throwing storage infrastructure optimization out the window. Flash is making performance management and tuning a lost art form. The problem is performance management is still very necessary. Without it IT is buying more flash than they potentially need. It is also creating a massive challenge for itself when the day comes where flash doesn’t solve all the organization’s performance challenges.
When flash was expensive, organizations spent hours determining if they should invest in flash and hours more determining which applications would go on flash. Because of the expense of flash, organizations explored every other option prior to making the all-flash plunge.
The Downside of Optimization
Some methods of extracting extra performance definitely don’t make sense anymore. For example, prior to the accessibility of enterprise flash storage, storage infrastructure tuning included using dozens if not hundreds of hard disks in an array. The more spindles the faster the response time. To that, IT planners implemented a process called short stroking, where only the outer platter of the hard disk was formatted, wasting more than half the capacity but improving response time.
Beyond trying to extract more performance from a slow technology like hard disk drives, other optimization steps also create more problems than they were worth. An example is database optimization just for the sake of working around hard disk performance limitations. These steps typically involved dividing up the database so it was smaller and could be housed on two separate storage systems.
But there are parts of the infrastructure that still need optimization, even after flash. And most of these optimizations will make the flash investment more reliable and have a longer useful life. Imagine getting 10 years of useful service instead of five.
There are parts of performance optimization that still make sense, even in the flash age. In fact, flash actually exposes the weaknesses. One of the first areas to look at is the network. As a general rule of thumb, implementing a new storage system should cause CPU utilization to rise dramatically. If after implementing flash you don’t see a corresponding rise in CPU utilization, especially as you add additional workloads to the environment, then more than likely there is a networking issue. Instead of waiting on the storage system, the CPU is now waiting on the network.
Network optimization can come in many forms. Most obviously the organization needs to consider an upgrade in bandwidth. But the organization should also examine the network itself. Look for improperly set ports or inter-switch links (ISL).
Eventually, all performance optimization ends with database tuning. The first step is to tune SQL queries. Can an index be added to help reduce query time. Are there queries automatically running that are no longer needed? Another important step is to monitor indexes knowing when indexes should be dropped and then dropping them can be critical to performance.
With all these recommendations why do we suggest organizations actually buy a flash array first? Storage Switzerland is suddenly starting to sell flash arrays (or anything for that matter), that’s not the motivation. The reason for a flash-first mentality is it buys the organization time. Most performance problems need urgent remediation, and for the overwhelming majority of situations a flash array provides that. Optimization can and should occur after the flash array is purchased and before the next performance problem rears its ugly head.
To learn more about optimizing performance before and after a flash investment, check-out our on-demand webinar, “Optimizing Flash Storage for SQL Databases“.