Soft real-time applications require timely delivery of messages conforming to the soft real-time constraints. Satisfying such requirements is a complex task both due to the volatile nature of distributed environments, as well as due to numerous domain-specific factors that affect message latency. Prompt detection of the root-cause of excessive message delay allows a distributed system to react accordingly. This may significantly improve compliance with the required timeliness constraints. In this work, we present a novel approach for distributed performance monitoring of soft-real time distributed systems. We propose to employ recent distributed algorithms from the statistical signal processing and learning domains, and to utilize them in a different context of online performance monitoring and root-cause analysis, for pinpointing the reasons for violation of performance requirements. Our approach is general and can be used for monitoring of any distributed system, and is not limited to the soft real-time domain. We have implemented the proposed framework in TransFab, an IBM prototype of soft real-time messaging fabric. In addition to root-cause analysis, the framework includes facilities to resolve resource allocation problems, such as memory and bandwidth deficiency. The experiments demonstrate that the system can identify and resolve latency problems in a timely fashion.