The architecture for the majority of network management systems (NMS) rely on a central polling system that is required to scale with the number of devices and services on those devices being monitored. This requires a sophisticated polling system and often these are proprietary and cost an arm and a leg. A number of these systems are listed in my Alpha to Zulu guide of monitoring. As there is no intelligence in the device being polled the polls themselves are numerous and not optimized. A large amount of raw metrics is extracted and then needs to be processed to generate a valid metric to be stored, analyzed and graphed.
Read the article over at LinkedIn here.