Comprehensive Application Performance Monitoring for E‑commerce Peak Traffic

Author : Wood, CTO of Tingyun, with over 15 years of experience in internet and software services, focusing on application performance monitoring and optimization.

As the annual Double‑11 shopping festival approaches, engineers must address massive concurrent traffic, handling massive data, network latency, and load balancing to ensure a fast, stable, and flashy user experience.

A complete Application Performance Management (APM) solution is essential for early problem detection and rapid resolution, keeping downtime at bay.

Monitoring Content

Four layers are described:

Server resource monitoring: disk space, CPU usage, memory, I/O, network traffic.

Network link monitoring: internal network connectivity, routing, and external user network status such as DNS and CDN quality.

Application layer monitoring: web containers, databases, NoSQL, mobile apps—metrics like cache hit rate, JVM health, QPS, response time, request status, queue length, slow SQL, Memcache/Redis status, app launch rate, crash reports.

Business layer monitoring: key business throughput and logic smoothness—order count, new registrations, complaints, queue lengths, and user experience metrics.

Monitoring Methods

Three typical ways to build a monitoring platform:

Combine open‑source software with shell scripts.

Develop custom monitoring tools and platform.

Purchase third‑party monitoring services.

All approaches usually adopt a distributed architecture to avoid high communication and computation costs of centralized processing.

Server Resource Monitoring Methods

Beyond OS commands (e.g., top, free), major e‑commerce companies like Alibaba and JD.com open‑source tools such as Tsar to collect system and application metrics, which can be fed into remote databases or visualized with Nagios/Cacti.

Network Link Monitoring Methods

Internal network status is monitored similarly to server resources using scripts and tools (Ping, Tracert) integrated into Nagios/Cacti. External network monitoring can be done by deploying custom probes (e.g., Alibaba’s Alibench) or purchasing services like Tingyun Network, which collect DNS, TCP, DOM parsing, and rendering data from browsers.

Application Layer Monitoring Methods

Typical methods include log analysis via scripts, using built‑in statistics commands, and collecting metrics such as QPS, response time, errors, JVM status, slow SQL logs, and cache health. Third‑party APM services (e.g., Tingyun Server) automatically trace HTTP requests and code‑level performance. Mobile app monitoring requires embedding SDKs to gather interaction, launch, and crash data.

Business Layer Monitoring Methods

Key business metrics are exposed via custom APIs and visualized in Nagios/Cacti. Full business‑process monitoring often relies on third‑party tools that record and replay user flows to capture performance and error information.

Case Study

Figure 2 shows a monitoring curve of an e‑commerce checkout flow. On March 3, availability dropped to ~65 %, triggering an alarm (threshold 90 %). Within five minutes, alerts were sent, engineers identified a JavaScript error caused by an incomplete compatibility test of a new system version, and the issue was fixed before the next traffic peak.

Conclusion

Effective monitoring enables rapid detection of user‑facing problems during e‑commerce peak periods, greatly improving shopping experience and preventing financial loss; in short, “monitoring is omnipresent in e‑commerce peak architectures.”