Internet-based multimedia applications (e.g., voice-over-IP, instant messaging, and video conferencing) are continuing to grow in importance as more people depend on such applications for personal and professional communications. Although performance is almost always a concern with multimedia systems that must satisfy quality-of-service (QoS) constraints, security is also a major requirement given the increasing criticality of such applications. For example, businesses might depend on Internet telephony to reach customers while governments might depend on video streaming to disseminate information. For distributed multimedia services, in addition to the traditional security properties (confidentiality, integrity, and availability), accountability is also important to complement perimeter defenses. Accounting for user actions within the system enables the development of higher-level security services.This dissertation will present the Kantoku framework, which includes several different accounting mechanisms for different environments ranging from single servers to server clusters to computing clouds. Our framework focuses on distributed multimedia services deployed in such environments.In particular, we will show how our Kantoku framework can be used to address the problem of transaction state overload on multimedia servers.The primary attack that we consider throughout this dissertation is a novel denial-of-service attack that intentionally induces transaction state overload at multimedia servers. We refer to this attack as a Ringing-based denial-of-service attack. Unlike many denial-of-service attacks that rely on an increase in the incoming traffic rate, a Ringing-based denial-of-service attack only requires an increase in the transaction durations while the incoming traffic rate does not change. Such attacks cannot be detected by protection mechanisms that monitor the network traffic for anomalies. After briefly explaining some background information on the Session Initiation Protocol, we characterize transaction state accumulation during a Ringing-based denial-of-service attack both mathematically and empirically through extensive experiments.The first solution to preventing transaction state overload that we present is a family of early ii termination algorithms that selectively terminate transactions suspected of causing transaction state overload in multimedia servers. This protection mechanism relies on per-transaction accounting at a single server. The three algorithms that we developed are thoroughly evaluated in the context of Ringing-based denial-of-service attacks with experiments on a local testbed.As an alternative to early termination, we also developed two admission control algorithms that selectively reject transaction requests from users suspected of consuming more than their fair share of transaction resources among a group of multimedia servers. This protection mechanism relies on per-client accounting across a cluster of multimedia servers. The two algorithms that we developed are t...