Next Generation Intelligent APM: Pain Points, Trends and Solutions
?
3 likes?677 views
This document discusses the pain points of traditional application performance monitoring (APM) and trends toward next generation intelligent APM. It outlines four main pain points: increasing app complexity, needing to investigate multiple data sources across teams, data volumes being too large to effectively monitor, and heterogeneous APM data types. It then presents a vision for next generation APM that leverages unified monitoring, machine learning to extract insights from massive APM data, and moves from reactive to proactive monitoring through predictive analytics and anomaly detection. Examples of how machine learning could help with tasks like log analysis, smart alerting, root cause analysis and data summarization are provided.
Next Generation Intelligent APM: Pain Points, Trends and Solutions
- 2. Application Support Application Development IT Operations DevOps Release Application Owner ? an APM Company ? a Leader in Gartner's Magic Quadrant ? highest scores in every APM Use Case (Gartner, 2016)
- 3. Why APM?
- 8. 8/8/2016 Delta Airlines System Outage ? Flights canceled: 1,000+ ? Passengers Affected: hundreds of thousands ? Days to recover: 4+ ? Loss: millions of $$$
- 18. ? mobile ? browser ? web server ? middleware ? database 10+ tools! ? storage ? cloud ? network ? log ? IoT
- 33. We have metrics / time series: ? CPU usage ? Java JVM metrics ? network latency ? daily active user ? hourly sales amount ? ¡
- 34. We have categorical data: ? node name ? service name ? region ID ? user IP ? payment method ? ¡
- 36. We have graphs:
- 37. We have unstructured data: ? logs ? error messages ? stack traces ? ¡
- 43. See Act Know
- 45. Unified Monitoring one scalable data platform, end-to-end visibility Application Performance Management Database ? Monitoring Synthetic ? Monitoring Browser Real-User? Monitoring Mobile Real-User? Monitoring Server? Monitoring Flexible ? Deployment SaaS On-Prem Unified Monitoring
- 46. Auto Discovery automatic distributed flow discovery for micro-services Application Performance Management Database ? Monitoring Synthetic ? Monitoring Browser Real-User? Monitoring Mobile Real-User? Monitoring Server? Monitoring Flexible ? Deployment SaaS On-Prem Database API Server ApplicationWeb? Server Business? Transaction Browser Mobile Unified Monitoring
- 48. See Act Know
- 52. See Act Know
- 55. Writing regex: ? requires domain knowledge ? tedious ? error-prone
- 59. Use machine learning: ? Automatic baseline ? Dynamics threshold ? Seasonality analysis Techniques: ? auto regression ? PCA ? wavelet decomposition
- 60. See Act Know
- 67. APM Applications: ? relevant signal recommendation ? time series modeling ? metric prediction ? auto regex inference ? unstructured data clustering ? anomaly detection ? graph optimization ? ... ML frequently applied to APM: ? clustering ? neural net / deep learning ? classification ? ranking ? recommendation ? graph/network analysis ? prediction ? ...
- 69. Recap APM pain points: ? App complexity is exploding ? Investigation of multiple sources ? Data is too big ? Data is heterogeneous The next generation APM: ? Unified monitoring ? Data-driven intelligent APM (see ¡ú act ¡ú know) ? reactive ¡ú real-time ¡ú proactive
- 70. Thank You