How intelligent is your monitoring agent?
The agent should not do too much processing locally to ensure minimal impact to application performance by utilizing the smallest CPU and memory footprint possible. On the other hand, offloading some processing to the agent results in less network traffic and more scalability from the monitoring Mgmt Server.
The less intelligent agent just collects as many metrics as possible and throws them at the Mgmt Server. This generates a lot of network traffic and loads the Mgmt Server with unnecessary processing, which means you need a lot of hardware to scale. There is no benefit in sending millions of metrics to the Mgmt Server when Business Transactions are performing with in normal limits; this just degrades the signal-to-noise ratio and results in extra overhead. Monitoring isn’t about collecting everything–it’s about collecting what is relevant.
James Bond 007
An intelligent agent like AppDynamics does some local processing while maintaining a low profile. It only sends pertinent information back to the Mgmt Server, providing a high signal-to-noise ratio and ensuring low overhead. The agent baselines all Business Transactions locally and only sends back diagnostic information when performance is non optimal. This ensures that poor Business Transaction performance is detected immediately but network chatter is kept to a minimum. The high signal-to-noise ratio means that the Mgmt Server only has to deal with information from Business Transactions outside of normal resulting in high scalability for large enterprise environments; currently, the largest deployment on one Mgmt Server is over 4500 JVMs.
Who do you want looking after your applications–Mr Bean or 007?