Analysis of Monitoring Structures for Network-on-Chip: A Distributed Approach

Analysis of Monitoring Structures for Network-on-Chip: A Distributed Approach

Ville Rantala (University of Turku, Finland), Teijo Lehtonen (University of Turku, Finland), Pasi Liljeberg (University of Turku, Finland) and Juha Plosila (University of Turku, Finland)
DOI: 10.4018/978-1-4666-2776-5.ch008


Monitoring services are essential for advanced, reliable NoC systems. They should support traffic management, system reconfiguration and fault detection to enable optimal performance and reliability of the system. The paper presents a thorough description of NoC monitoring structures and studies earlier works. A distributed monitoring structure is proposed and compared against the structures presented in previous works. The proposed distributed network monitoring system does not require centralized control, is fully scalable and does not cause significant traffic overhead to the network. The distributed structure is in line with the scalability and flexibility of the NoC paradigm. The paper studies the monitoring structure features and analyzes traffic overhead, monitoring data diffusion, cost and performance. The advantages of distributed monitoring are found evident and the limitations of the structure are discussed.
Chapter Preview

Monitoring In Noc

An NoC monitoring system has the same main requirements as the NoC itself; it should be flexible, scalable and it should also be capable of real-time operation (Ciordas, Basten, Radulescu, Goossens, & Meerbergen, 2005). Scalability and flexibility ensure that a monitoring system can be used in different sized NoCs without a time consuming redesign process. A monitoring system itself should also be fault tolerant.

A network monitoring structure consists of monitoring units (monitors) and probes as well as communication resources, or wires, to connect these components to each other and to the NoC, as illustrated in Figure 1. The probes are connected to network components, e.g., routers, network interfaces or links, whose functionality they observe and deliver the observed data to the monitoring units. If there are dedicated resources for communication between the probes and the monitors, the probes can be controlled from the monitors. Otherwise the probes have to be more autonomous to be able to communicate over the shared packet switched network. Monitors collect the monitoring data, exchange the data with other monitors and provide information for network components which can use it to reconfigure their operation. The system diagnostics can share the resources of the traffic management but may also require dedicated components, e.g., probes which are attached to the computational resources and centralized control devices which are not necessarily required in the traffic management.

Figure 1.

Main components of network monitoring systems. Network components to be monitored are e.g. routers, links or network interfaces.


Complete Chapter List

Search this Book: