Modeling High-Performance Wormhole NoCs for Critical Real-Time Embedded Systems

Abstract

Abstract-Manycore chips are a promising computing platform to cope with the increasing performance needs of critical realtime embedded systems (CRTES). However, manycores adoption by CRTES industry requires understanding task's timing behavior when their requests use manycore's network-on-chip (NoC) to access hardware shared resources. This paper analyzes the contention in wormhole-based NoC (wNoC) designs -widely implemented in the high-performance domain -for which we introduce a new metric: worst-contention delay (WCD) that captures wNoC impact on worst-case execution time (WCET) in a tighter manner than the existing metric, worst-case traversal time (WCTT). Moreover, we provide an analytical model of the WCD that requests can suffer in a wNoC and we validate it against wNoC designs resembling those in the Tilera-Gx36 and the Intel-SCC 48-core processors. Building on top of our WCD analytical model, we analyze the impact on WCD that different design parameters such as the number of virtual channels, and we make a set of recommendations on what wNoC setups to use in the context of CRTES.

Boosting Guaranteed Performance in Wormhole NoCs with Probabilistic Timing Analysis

“…Routers are pipelined and consist of 4 stages: input buffer, routing, switch allocation, and crossbar traversal. In line with other works [19], [18] in all wNoC setups we use single-flit packets only to improve performance guarantees. The number of VCs is 1.…”

“…Upper-bounding latency deterministically only requires forcing all wNoC requests to experience the worst-possible delay [19]. To illustrate probabilistic upperbounding, let us assume a hardware resource whose analysistime latency can be 1 or 2 cycles with the same probability: etd a =< (1, 2) , (0.5, 0.5) > where the first vector corresponds to the different latencies and the second to their associated probabilities.…”

“…In a wNoC setup with all-to-all traffic the worst contention situation can be reproduced by considering that the flows contending for the resources with the flow of the core under analysis (F i ) are all worst-possible destination flows [19]. The destination of contenders' flows is chosen such that it causes the worst contention to F i packets (i.e.…”

“…It is important to mention that reducing contention by reducing the number of requests in the network is suitable for probabilistic wNoCs because, in such designs, requests interleave probabilistically and therefore, worst-case alignment of flows and arbitration decisions do not need to be accounted for systematically (as opposed to the case of time-deterministic wNoCs). In other words, already proposed techniques for reducing contention in time-deterministic wNoCs, such as injection throttling [24], cannot obtain tighter WCET estimates [19].…”

“…Unlike buses or other existing centralized network architectures, wNoCs perform the arbitration of communication flows in a distributed manner, which severely complicates the derivation of request contention bounds as required in real-time domains. In this line, some works show that, while reliable contention upper bounds can be provided for Commercial off-the-shelf (COTS) wNoCs [21] [19] [18], those bounds are pessimistic, preventing an efficient use of high-performance wNoCs for mixed-criticality real-time embedded systems (RTES).…”