Designing an Operations Monitoring Platform: Tools & Best Practices

Operations professionals often say, "No monitoring, no operations," emphasizing that monitoring is the "third eye" of the field. Without it, both basic and business operations are blind, making monitoring the foundation of the profession.

In the era of DevOps, data‑driven monitoring becomes essential; with sufficient data, operations can speak with facts rather than blame.

This article examines how to select monitoring tools and design a unified monitoring platform, useful for newcomers and seasoned engineers alike.

Common Operations Monitoring Tools

There are many monitoring tools, each with distinct characteristics.

Cacti

Cacti is a PHP‑based network traffic monitoring and graphing tool that uses SNMP and RRDTool. It visualizes performance trends but lacks distributed support, alerting, and produces less attractive graphs.

Nagios

Nagios is an open‑source monitoring solution for hosts, networks, and devices, offering robust alerting via email or SMS. However, it lacks strong data collection, has limited graphing, and configuration is text‑based without a web UI.

Its strength lies in alerting, but it suffers from cumbersome host addition and error‑prone manual configs.

Zabbix

Zabbix provides a web‑based, distributed monitoring solution with strong notification mechanisms. It consists of a server and optional agents, supporting SNMP, ping, port checks, and runs on many platforms.

Zabbix addresses Cacti’s alerting gap and Nagios’s web‑config limitation, supporting distributed deployment, though it can be resource‑intensive at large scale.

Ganglia

Designed for HPC clusters, Ganglia is a lightweight, distributed monitoring system that collects CPU, memory, disk, I/O, and network metrics via gmond agents, aggregates them with gmetad, and visualizes data through a PHP UI.

Its low overhead complements Zabbix’s higher resource usage, making it suitable for large‑scale data platform monitoring.

Centreon

Centreon builds on a Nagios‑like engine, adding a web UI for host configuration, distributed monitoring, and integration with Ganglia for unified data presentation.

Prometheus

Prometheus is an open‑source monitoring and alerting framework suited for both hardware metrics and dynamic micro‑service environments, offering powerful multidimensional data collection and a query language.

Grafana

Grafana is an open‑source visualization suite that can display data from many sources (Graphite, InfluxDB, OpenTSDB, Prometheus, Elasticsearch, CloudWatch, KairosDB) with attractive dashboards.

Unified Operations Monitoring Platform Design

A monitoring platform must integrate data collection, visualization, extraction, alert rule configuration, event generation, and user management into a cohesive system.

The architecture consists of six layers:

Data Collection Layer : Gathers network, system, application, and database metrics, normalizes and stores them. Data Presentation Layer : Web UI that visualizes collected data as charts, aiding troubleshooting. Data Extraction Layer : Filters and formats data for the alerting module. Alert Rule Configuration Layer : Defines thresholds, contacts, and notification methods. Alert Event Generation Layer : Records alerts, stores them, and produces reports. User Management Layer : Provides multi‑user, role‑based access to the web UI.

The platform is divided into three functional modules: data collection, data extraction, and monitoring/alerting, each supporting various tools (e.g., Cacti, Ganglia for collection; Nagios, Centreon for alerting).

Enterprise Monitoring Platform Selection

SMBs : Zabbix offers an all‑in‑one solution with quick onboarding, though it may require distributed deployment and HA for large host counts.

Large Internet Companies : A combination of Ganglia (lightweight data collection) and Centreon (rich UI and alerting) provides scalable, low‑overhead monitoring.

Evolution of Our Monitoring Platform

Different stages of machine count dictate platform design:

Less than 100 hosts

Simple deployment, stability, basic alerting via email/SMS; tools like Nagios, Cacti, Zabbix, Ganglia are suitable.

200–1000 hosts

Introduce classification of monitoring data, full‑coverage monitoring, multi‑channel alerts, and optimize alert policies to reduce noise.

Over 1000 hosts

Address alert latency, single‑point failures, and business‑logic monitoring by adopting distributed proxies, HA setups, and custom extensions (e.g., integrating Ganglia for low‑overhead collection with Zabbix for business metrics).

Ultimately, a well‑designed monitoring platform empowers operations teams to proactively manage infrastructure and services.

Author: 南非蚂蚁 Source: http://blog.51cto.com/ixdba/2310782