Solid-state storage can
Many storage products now offer some kind of integrated solid-state storage, such as all-flash arrays, intelligent dynamic auto-tiering, hybrid arrays or solid-state storage added to existing traditional controller-based HDD storage arrays.
Using this high-performance technology is great for accelerating a single application to a whole IT infrastructure, which can reduce the amount of infrastructure needed to meet SLAs and save in Capex and Opex if the infrastructure is well-architected.
But is using solid state in too many areas of an infrastructure causing diminished returns on performance?
Problems where once there were none
When a storage bottleneck is alleviated, this can create new performance bottlenecks or move bottlenecks to other resources. For example, a popular economical solid-state solution being used today is server-side solid-state storage coupled with a software caching engine using write-through caching with optimized writes sent immediately to the back-end storage.
But is the back-end storage configured correctly to handle these optimized writes? If not, this could create a new bottleneck with the back-end storage controller's write cache and even possibly a bigger performance bottleneck for write misses to hard disk drives if there is no solid-state storage tier.
An initial indicator if writes are properly optimized is to monitor response times to ensure they stay under SLA and workload performance threshold requirements. On the positive side, the server-side solid-state storage solution can dramatically reduce the amount of read cache needed for storage controllers since the server-side caching engine is handling all reads. This frees the read controller cache for use with non-accelerated workloads. In addition, a high read cache hit rate on a storage controller is usually an indicator of good performance, but this can be misinterpreted in this example since all of the reads are cached on the server, and the read cache on the storage controller is no longer used.
Other performance-impeding considerations
In clustered virtual environments using advanced features, such as VMware's vMotion, DRS or Storage DRS, with server-side solid-state storage and a supported caching engine, there are particular performance considerations to deal with.
For example, the cache needs to be "rewarmed" on a new host after a virtual machine has been migrated from one host to another host. While data can be copied to solid-state storage from one host to another very quickly, getting the right data back into cache can take time, since that data has to be accessed and then reevaluated for cache worthiness. This means an application that was likely accustomed to high performance on solid state may perform at a hard disk drive level until the cache re-warming process is complete. This can take longer than expected, depending on the workload intensity of the cluster.
Another performance bottleneck to consider for clustered environments using server-side solid-state storage is the inter-nodal traffic that goes on among the distributed PCIe cache cards in each server of the cluster. This introduces its own overhead that should be monitored and managed.
Many storage systems have been redesigned to handle large volumes of I/O requests to maximize the use of integrated solid-state storage. Traditional storage controllers are being redesigned not only to handle the additional performance of solid state but also the unique characteristics of solid-state behavior versus the mechanics of hard-disk drives. All-flash arrays and flash controllers offer the higher performance needed to meet the increased demands of today's IT requirements.
I/O-intensive applications and environments will introduce new storage bottlenecks faster than the average application environment. New storage systems are able to handle a larger volume of I/O requests with solid-state storage, which eliminates many storage bottlenecks traditionally seen in HDD storage. As the use of solid-state storage matures, some IT departments are seeing a growing number for network performance bottlenecks in their infrastructures. The next stop for application acceleration will be in the network infrastructure.
About the author:
Leah Schoeb is a senior partner at Boulder, Colo.-based Evaluator Group.
This was first published in August 2013