This article can also be found in the Premium Editorial Download "Storage magazine: Solid-state adds VROOM to virtual desktops."
Download it now to read this article plus other related content.
Dealing with the database. The final database performance condition involves situations where the whole database needs to be accelerated, not just its index and log files. This can emerge as an issue because the bulk of the queries or the transaction rate exceeds the ability of the storage system to keep pace with the environment. There are two options in these situations: moving the entire database environment to a faster storage system or using a flash-based caching technology.
Opting for a new, faster storage system doesn’t necessarily mean a move to flash-based storage. Contemporary mechanical storage can still turn out ample IOPS when properly tuned. The legacy practice of short-stroking drives can also be avoided. As described above, short stroking means drives are formatted at one-half or one-third of their normal capacity by using only the outer tracks or cylinders of each disk platter. This improves response time because the linear speed of the disk track moving past the drive head is faster on the outer cylinders. Using less than each drive’s full capacity increases overall drive count, which can also improve performance by allowing more queued storage requests to be simultaneously addressed. Unfortunately, this practice is also expensive, wasteful and a significant draw on power resources when applied to the entire database.
Most current enterprise storage systems can provide
Another option is to use a flash-based caching technology. These products can be installed in the server, on the network or on the storage system itself. Each location has its advantages and disadvantages. The server typically provides the best performance as the cache is often a PCI Express (PCIe) SSD and has direct access to the server’s CPU. However, those caches typically only cache reads, especially in an environment where the database is shared, due to concerns about maintaining “coherency” between the cache and the primary storage it’s supporting. Network caches can typically do both read and writes, and work across a variety of storage systems. They’re available for NFS or Fibre Channel (FC)-attached environments. The challenge is the added network latency. Finally, storage system-based caching is also available but it only works on the storage system in which it’s installed and isn’t universally available across a variety of vendor storage systems.
Start your engines…
Improving database storage performance is a process. In most environments, there’ll always be an area that requires some tuning. The key is to tune it to the point where today’s performance demand is met and deal with tomorrow’s performance issues when they arise. Technology will continue to evolve and the storage manager’s toolkit will become more diverse and effective over time. A step-by-step approach also allows a minimal amount of your budget to be spent. A database storage performance problem doesn’t always mean ripping and replacing the storage system.
BIO: George Crump is president of Storage Switzerland, an IT analyst firm focused on storage and virtualization.
This was first published in July 2012