APM vendors need to focus on monitoring storage I/O latency as part of end-to-end transaction analysisThere is a lot of conflicting information about what constitutes a great application performance management (APM) solution. The reality is that the right solution depends on your needs. And it also depends on where you can get the most bang for your buck. In a Gartner research note titled APM Technologies Must Monitor Storage I/O Latency, Will Cappelli makes a compelling case for why APM vendors must pay attention to storage I/O latency as part of end-to-end transaction analysis.
Cappelli points out that traditional application performance management (APM) vendors do not break out storage I/O latency from overall database response time. This approach has worked adequately in the past because storage tended to be local to the databases and was not a large part of the database response time. However, according to Gartner's research, during the past two years, enterprise architectures have shifted largely to non-local storage media, such as network-attached storage (NAS) and storage area networks (SAN). The increased flexibility associated with these storage solutions comes at a cost: a much more significant I/O latency component.
Traditional APM vendors, such as Compuware and Visual Network Systems, analyze performance from the end user up to the database layer, but not beyond. Any storage latency associated with database lookups is measured as part of the database wait time. Separate storage analysis tools are available directly from the storage vendors, such as IBM, HP, EMC and CommVault; however, these tools are limited only to analyzing storage environments. According to Cappelli, this limitation presents a problem, because "the ability to understand the contribution of storage I/O to end-to-end latency typically requires the manual collecting and correlating of data from database performance monitoring tools … on the one hand, and storage management tools … on the other."
The ExtraHop Application Delivery Assurance system understands the central role storage plays in end-to-end application performance management. In addition to timing database wait time, the ExtraHop system analyzes the latency of the back-end storage components. Analysis is performed without any agents, via passive analysis of network traffic between database and storage systems. The solution supports both NAS protocols, such as CIFS and NFS, as well as increasingly important SAN protocols, such as iSCSI. A variety of advanced storage metrics go beyond latency analysis to assist with hard-storage failure conditions affecting application delivery. Finally, file-level transaction analysis enables storage administrators to monitor and optimize resource usage and plan for future capacity.
With recent shifts in enterprise architectures, each new layer of complexity introduces additional performance burdens. Businesses are becoming overwhelmed by problems such as storage I/O latency and hindered by the inadequacies of APM products that monitor each tier separately, failing to provide a central view of their environment in real time. The ExtraHop system delivers much-needed end-to-end visibility into not only storage, as discussed by Cappelli, but all tiers, including network, web, and database.
We'd love to hear how your IT team monitors the storage tier.