NoC Reconfiguration for CMP Virtualization

Abstract

At NoC level, the traffic interferences can be drastically reduced by using virtualization mechanisms. An effective strategy to virtualize a NoC consists in dividing the network in different partitions, each one serving different applications and traffic flows. In this paper, we propose a NoC reconfiguration mechanism to support NoC virtualization under real scenarios. Dynamic reassignment of network resources to different partitions is allowed in order to NoC dynamically adapts to application needs. Evaluation results show a good behavior of CMP virtualization.

Mixed-criticality NoC partitioning based on the NoCDepend dependability technique

“…They are suitable for efficient multicast communication only, but not for protection of spacial domains within the network. The authors of [8] introduce a domain partitioning approach which is spacial and and based on logic-based distributed routing (LBDR), which allows the separation of local domains by canceling the connectivity bits to adjacent routers. This method has the potential to hardware-protect domains, but is limited to the LBDR mechanism, which allows local circumvention of faulty routing resources in the NoC.…”

“…Since to the best of our knowledge only [8], [11] provides a comparable real configurable hardware protection, it is used as comparison benchmark. While the method in [8] is limited to LBDR-based routing and concave shapes of domains, the approach presented in this paper can be combined with any routing algorithm and works for basically any shape of criticality domains (as long as these shapes can be loss-less represented by the amount of available rectangle entries in every outgoing router link).…”

“…While the method in [8] is limited to LBDR-based routing and concave shapes of domains, the approach presented in this paper can be combined with any routing algorithm and works for basically any shape of criticality domains (as long as these shapes can be loss-less represented by the amount of available rectangle entries in every outgoing router link). Both approaches can gain the partitioning without additional effort on top of the dependability management.…”

“…Both approaches can gain the partitioning without additional effort on top of the dependability management. In contrast to [8] the NoCDepend fault-tolerance technique can tolerate any amount of network faults and allows circumvention of faulty regions, as long as allowed by the deadlock-free turn model and the routing graph is still connected. Additionally the presented method provides gateways for inputs and outputs to/from critical regions and any reconfiguration of the amount, size and shape of critical regions is possible.…”