Performance Measurement of the Scalable Sensing Service (S) on the ProtoGENI Control Framework

Accurate performance evaluation of any deployed system is important to its user community. The distributed nature of network measurement infrastructures escalates the complexity of this performance evaluation. In this work, we evaluate the performance of a measurement service, the S system [3]. We measure the performance in terms of resource utilization and response times under increasing measurement load. By response time, we mean the time between issuing a measurement request to the time we start obtaining measurement results. Resource consumption of the measurement service is also an important attribute to measure, since it can limit the scalability of the system. 1 Overview of S S3 [1] is a service for real-time and configurable monitoring of large networked systems. Its main components are the sensor pods, the sensing information management system (SIMS), and scalable inference engines. Sensor pods execute light-weight network measurement tools for sensing network path properties between any two end-nodes. SIMS is the component responsible for initiating the sensing (measurement) commands and later collecting sensed information from the sensor pods. The main purpose of inference engines is to leverage sensing tools on the sensor pods for more efficient measurement. The S3 system we are profiling in this work is the prototype deployed on the ProtoGENI infrastructure [2]. We evaluate S3 performance with respect to its resource consumption on both the end-nodes and the network, i.e., we will compute the entire system footprint. We will also evaluate response times for performing different operations with S3.