Application performance management (APM) is critical for a web-facing business as any performance degradation or instability will impair work efficiency and result in revenue loss. According to Amazon, every 100 millisecond system latency increase causes 1% sale reduction. Google finds the search volume is reduced by 20% with each additional 0.5 second query response delay. ITRI’s Agentless Application Performance Management (AAPM) is a software technology that can accurately pinpoint the root cause of the performance problems experienced by large-scale distributed applications, without installing any agent or modifying the applications. This technology requires no deployment effort and is maintenance-free; moreover, it is applicable for application software whose source code is not available.
ITRI’s AAPM adopts an agentless approach to achieve application performance optimization.
Typical APM systems modify programs to retrieve application behaviors and require installing an agent or recompile with instrumented libraries, which becomes burdensome to the data center operation and maintenance and also affects system stability. AAPM is a hypervisor-based solution using an agentless approach. It shows better monitoring and analysis capabilities and can achieve the following advantages:
Comparison of ITRI’s AAPM and a typical approach.
AAPM features detailed dependency mapping and intelligent problem detection. It gathers real time information on the incoming and outgoing traffic of the virtual machines, and uses this information dependency map to discover distributed application topology as well as detect possible anomalies. Given a set of distributed applications running on a cloud data center, AAPM is able to automatically determine the set of services provided by each application, enumerate the set of processes/threads involved in each of these applications, and reconstruct the request/response relationships among them into a dependency graph. For each service in an application, AAPM measures and reports the request-response time of each edge on the service’s corresponding trajectory. These real-time fine-grained measurements allow an application administrator to quickly resolve the core issues associated with observed or reported performance problems. Moreover, AAPM offers the above measurements without requiring any application modification or re-compilation.
AAPM, consisting of a packet logger, an introspector, and an analytic engine, supports socket based and message queue based applications. The system can optimize performance management for cloud services, big data platforms, and networks. This technology enables active application performance optimization, supports visualization interfaces, and facilitates flexible analytical tool deployment and update. Coupled with big data analysis technology for in-depth packet analysis, AAPM can also support advanced network performance monitoring and analysis.