Planeten Wach Hund Netz: Instrumenting infrastructure for Planetlab
The distributed and highly dynamic nature of Peer-to-Peer (P2P) systems can make instrumentation and development difficult. Instrumenting a system aids development by producing debugging and status information. Moreover, collection of data from such systems has the potential to produce vast amounts of data that preclude the use of approaches that collect unfiltered logs from many participating nodes. My research investigates the dynamic generation of trees that permit the efficient aggregation of data extracted from nodes by enabling the use of parallel prefix computations on P2P systems. These automatically structured parallel prefix operations are used to permit the efficient collection of instrumentation data. This paper describes PlanetenWachHundNetz (PWHN), an integrated instrumentation toolkit for distributed systems that uses the dynamically organized parallel prefix mechanisms I investigated. Like Google's MapReduce, PWHN utilizes user supplied programs to extract data from instrumented hosts and aggregate this data. In addition, convenient Graphical User Interfaces (GUIs) are provided to display collected data and generate plots. In order to evaluate the effectiveness of the parallel prefix aggregation mechanisms I investigated, PWHN supports both a centralized and distributed aggregation mode. PWHN's first (centralized) implementation has been useful for instrumentation and tuning of the live Fern installation on hundreds of PlanetLab nodes, but is not scalable. In contrast, the second implementation employs more scalable algorithms that dynamically construct data aggregation trees using P2P techniques and thus is potentially scalable to much larger deployments.
Lorenz-Meyer, Vitus, "Planeten Wach Hund Netz: Instrumenting infrastructure for Planetlab" (2007). ETD Collection for University of Texas, El Paso. AAI1441514.