One of the implementation details that is useful to understand is how the storage of performance data has progressed. Similar to how pg_stat_bgwriter was analyzed in the last chapter's examples, many of the interesting performance counters you'll want to monitor and then produce trend graphs of constant increase over time. The data that comes out of such a counter can be thought of as a series of (timestamp, count) samples. Two important questions to answer are where to store this series of points, and how to query them usefully.
The first generation of trend monitoring software was used to chart network performance data, typically collected using the Simple Network Management Protocol (SNMP) protocol that better ...