Master project: Scheduling for NoC Monitoring Service
Description
Future generation multimedia systems will very probably be Multiprocessor Systems on Chip (MP-SoCs) communicating via Networks on Chip (NoCs). NoCs are programmable on-chip interconnects organized in the form of an internet-like network of routers. NoCs can offer guaranteed throughput (GT) or best effort (BE) communication services.
Monitoring is the process of collecting information about the system behavior. Run-time monitoring collects the information at run-time allowing the understanding of the system’s behavior. Hardware event-based monitoring techniques have been little explored because of the huge amount of low-level data they generate. Nevertheless they proved to be low intrusive. NoC Monitoring Service comprises two main parts: one is the event capturing and the other is the off-chip event streaming. We have developed an efficient mechanism for event capturing and for adapting the monitoring demands (driven for example by debugging) on the fly. What is of interest for us right now is to investigate the means of on-chip transport of the monitored data to the debug port in a non-intrusive or predictably low-intrusive way.
Several approaches can be taken in account:
- The scheduling of the monitoring traffic on the network, on top of the existing functional traffic
- The scheduling of the monitoring traffic on dedicated monitoring connections, either GT or BE
- Taking advantage of the idle network cycles
- The overall traffic shaping in the network, where marginal network connections are seldom used
- Distributed versus centralized “monitoring nodes”
- Breakpointing the network and taking the data off
The project is be based on:
- existing AEthereal simulation environment (Philips Research NoC).
- existing simulation instrumentation techniques and configurable event service framework jointly developed by TUE and Philips Research.
Duration
August 2004 - May 2005
Student:
Supervisor:
Publications
|
