A lightweight throttling mechanism allows for dynamic control of access to resources in a distributed environment. Each request received by a server of a server group is parsed to determine tokens in the request, which are compared with designated rules to determine whether to process or reject the request based on usage data associated with an aspect of the request, the token values, and the rule(s) specified for the request. The receiving of each request can be broadcast to throttling components for each server such that the global state of the system is known to each server. The system then can monitor usage and dynamically throttle requests based on real time data in a distributed environment.