Communication in multicomputers with nonconvex faults

Abstract

A technique to enhance multicomputer routers for faulttolerant routing with modest increase in routing complexity and resource requirements is described. This method handles solid faults in meshes, which includes all convex faults and many practical nonconvex faults, for example, faults in the shape of L or T. As examples of the proposed method, adaptive and nonadaptive faulttolerant routing algorithms using four virtual channels per physical channel are described.

A Fully Adaptive Fault-Tolerant Routing Methodology Based on Intermediate Nodes

“…The presented throughput is the average of the individual results obtained when evaluating the 50 randomly generated fault combinations. 7 As can be observed, the throughput decreases as the number of faults in the network increases. However, the decrease in throughput, is very low.…”

“…Most of these fault-tolerant routing strategies require a significant amount of extra hardware resources (e.g., virtual channels) to route packets around faulty components depending on either the number of tolerated faults [9] or the number of dimensions in the topology [17]. Alternatively, there exist some fault-tolerant routing strategies that use none or a very small number of extra resources to handle failures at the expense of providing a lower fault-tolerance degree [9,14], disabling a certain number of healthy nodes (either in blocks (fault regions) [6,7] or individually [10,11]), preventing packets from being routed adaptively [15], or drastically increasing the latencies for some packets [19]. Moreover, when faults occur, link utilization may become significantly unbalanced when using those fault-tolerant routing strategies, thus leading to premature network saturation, and consequently, degrading network performance even more.…”

ARIADNE: Agnostic Reconfiguration in a Disconnected Network Environment

“…Other approaches place requirements on the configuration of faults, requiring them to be contained in convex regions [9,31], L or T shape regions [9], or faulty polygons [20]. Oftentimes the faulty region must be expanded to include non-faulty routers; as a result, non-faulty routers and links are disabled only to satisfy the algorithm's constraints.…”

“…A number of resilient routing algorithms that can be applied to any irregular topology (i.e., a topology that survived after a number of random faults in network links) has been proposed in this domain, including up*/down* (introduced in Autonet) [27], segmentbased routing [23], FX routing [26], L-turn [21], and smartrouting [10]. During reconfiguration, the surviving topology is communicated to a central node, which runs the reconreliability performance area bounded faults early work [12,16], VCs [17,18,29] flooding [6,24] limited n/a n/a pattern constraints convex [9,31], L or T [9], polygons [20] limited n/a n/a unbounded faults off-chip routing [10,21,23 figuration algorithm in software. Using global knowledge of the functional links, the software computes new routing tables and communicates them back to each node.…”

On constructing the minimum orthogonal convex polygon in 2-D faulty meshes

“…Most of the literature on fault-tolerant routing uses disjoint rectangular blocks ( [2], [3], [4], [7], [10], [18]) to model node faults (link faults can be treated as node faults) and to facilitate routing in 2-D meshes. First, a node labelling scheme that identifies nodes (faulty and non-faulty) that cause routing difficulties is defined and such nodes are called unsafe nodes.…”