A Scalable Framework for Wireless Distributed Computing

Fundamental tradeoff between computation and communication in distributed computing

“…Therefore, the coded data shuffling scheme of CDC can effectively reduce the communication loads of computation jobs on general data storage systems. This behavior that a random data placement achieves closeto-optimum performance has also been reported in [49] for a decentralized wireless distributed computing platform, and in [21] for a decentralized caching system.…”

“…In this case, we need to generalize our communication model to include more structured topologies, and develop coded shuffling strategies that account for (1) path lengths of shuffled data (2) congestion at links higher up in the topology; and (3) different link capacities and multicast-costs at different layers of network topology. We have made preliminary progress in [49] for a star topology (motivated by wireless edge computing), where nodes are connected via only one access point (or switch layer).…”

“…For an edge computing scenario where the mobile users upload the tasks to the edge nodes, and retrieve the computed results from the edge nodes, we have designed coded computing architectures in [54], [55], in which coded computations that are aware of the underlying physical-layer communication are performed at the edge nodes, achieving the minimum load of computation and the maximum spectral efficiency simultaneously. In [49], we have formulated a wireless distributed computing framework, in which a cluster of mobile users collaborate via an access point to simultaneously meet their computational needs. For this wireless computing platform, we exploited the coding techniques of CDC to achieve a scalable design such that the platform can accommodate an unlimited number of mobile users with a constant amount of bandwidth consumption.…”

Energy-Efficient Edge-Facilitated Wireless Collaborative Computing using Map-Reduce

“…Wireless distributed-computing (WDC) was already vastly studied in the literature, mostly from an information theoretic point of view [12]- [15]. These works mainly focus on coded distributed computing (CDC) and study the trade-off between the computation and communication loads incurred by the collaboration.…”

“…To make collaboration between the nodes possible, we also assume that the local data {d n } N n=1 were shared between all the nodes through the AP in a prior phase that we neglect in this work because D is assumed to be relatively small. In the absence of a reliable model for the energy consumption of coding operations and for the sake of tractability in order to lay out the foundations of the proposed collaborative-computing framework, coded Shuffling (i.e., CDC [12]- [15]) is not considered in this work. In this simplified Shuffle phase, each node n transmits the intermediate values v n,m = g n (d m , w n ) to node m via 1 In practice, ln should be an integer multiple of the size of the smallest possible division of w. In this work, we relax this practical consideration to avoid dealing with integer programming later on.…”