Exascale Computing Technology Challenges

By John M. Shalf, Sudip S. Dosanjh, John P. Morrison. 2010

443 citations recorded by the source corpus.

Subfields: Computer Networks and Communications, Hardware and Architecture

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Networks and Communications

In Hardware and Architecture

Citations in the corpus

Federated Optimization: Distributed Machine Learning for On-Device Intelligence (2016, background): “Compared with local computation on any single node, the cost of communication between nodes is much higher both in terms of speed and energy consumption [6, 89], introducing new computational challenges, not only for optimization procedures.”
Communication-Efficient Distributed Optimization of Self-Concordant Empirical Loss (2015, neutral): “O C ] 1 Ja”
Stochastic, Distributed and Federated Optimization for Machine Learning (2017, background): “The practical issue is that the time it takes to communicate between a processor and memory on the same node is normally many orders of magnitude smaller than the time needed for two nodes to communicate; similar conclusions hold for the energy required [165].”
DiSCO: Distributed Optimization for Self-Concordant Empirical Loss (2015, background): “Compared with local computation at each machine, the cost of inter-machine communication is much higher in terms of both speed/delay and energy consumption (e.g., Bekkerman et al., 2011; Shalf et al., 2011), thus it is often considered as the bottleneck for distributed computing.”
On the role of burst buffers in leadership-class storage systems (2012, background): “In the near future, these “leadership-class” systems are expected to have 100x to 1000x more compute nodes than today’s largest-scale HPC systems and require approximately 60 TiB/s of storage system bandwidth to drain application data bursts fast enough to meet checkpointing demands [12], [39].”
A review of predictive nonlinear theories for multiscale modeling of heterogeneous materials (2017, background): “Recently, the computational performance has increased through multi-level parallelism [231,249,250], while the per-CPU speed has remained nearly constant.”
Large-Scale and Extreme-Scale Computing with Stranded Green Power: Opportunities and Costs (2017, background): “With Dennard scaling long over [17], [18] power levels are growing rapidly: today’s 15-petaflop systems are >10MW [19], [20], announced nearexascale machines will exceed 15 MW [21], [22], with Exascale systems in 2023 projected >25MW.”
ZCCloud: Exploring Wasted Green Power for High-Performance Computing (2016, background): “With the end of Dennard scaling [1], [2], power has become an increasingly important concern for supercomputing centers.”