The Case of the Missing Supercomputer Performance: Achieving Optimal Performance on the 8,192 Processors of ASCI Q

By Fabrizio Petrini, D. Kerbyson, S. Pakin. 2003

565 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

Decentralized gradient methods: does topology matter? (2020, neutral): “The other was extracted from ASCI-Q super-computer traces (Petrini et al., 2003, Fig.”
Solving the Straggler Problem with Bounded Staleness (2013, methodology): “BSP provides a simple and easy-to-reason-about model for parallel computation, and can be easily applied to most iterative convergent algorithms.”
The Combinatorial BLAS: design, implementation, and applications (2011, background): “Run time variability of large-scale parallel codes, which can be due to various factors such as OS jitter (Petrini et al., 2003), is widely reported in the literature (Van Straalen et al., 2009).”
Robust Distributed Online Prediction (2010, background): “, [7]), let alone grid environments in which each node may have different capabilities.”
Automatic performance tuning of sparse matrix kernels (2003, background): “, showed factor of 2 performance improvements on a hydrodynamics application running on a large-scale parallel system (ASCI Q)—this improvement came solely from understanding and altering system software dynamics, requiring no changes to the application code [248].”
Addressing the straggler problem for iterative convergent parallel ML (2016, methodology): “Power-Law Pattern: Based on a real-world straggler pattern [42], this uses a power-law distribution [44] to model the time t for a worker to complete an iteration: p(t) ∝ tα , where α is the parameter that controls the “skewness” of the distribution.”
The RAGE radiation-hydrodynamic code (2008, methodology): “Details on ASCI Q (Compaq) performance as well as how SAGE was actually used in the optimization of the system performance is described in [9].”
Scalable molecular dynamics with NAMD (2005, background): “On larger machines, even occasional interruptions for normal operating system functions have been shown to degrade the performance of tightly coordinated parallel applications running on hundreds of processors [105].”
Arrakis (2014, background): “The High-Performance Computing field has long been interested in transferring data directly between user programs and the network, bypassing the OS which is often viewed as an undesirable source of “noise” [7, 23].”
Evaluating the performance of parallel subsurface simulators: An illustrative example with PFLOTRAN (2014, unknown): “…[ Mills et al ., 2010] when using many MPI processes, as the cost grows relatively quickly with the number of processes involved in the operation due to internode communication latency as well as load imbalance introduced by operating system ‘‘noise’’ [ Petrini et al ., 2003; Oral et al ., 2010].”