On the potential of significance-driven execution for energy-aware HPC

Gschwandtner, Philipp; Chalios, Charalambos; Nikolopoulos, Dimitrios S.; Vandierendonck, Hans; Fahringer, Thomas

doi:10.1007/s00450-014-0265-9

Cited by 10 publications

(15 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In other words, our use of residual norm in D1 allows us to use 1 as a significance filter for bit flips which may very often be insignificant. A wide range of bit flips which are insignificant in practice have been also observed in related work [20], [21]. Our solution has the implication that the two threads may diverge up to a certain point.…”

Section: B Detection and Correctionsupporting

confidence: 59%

TwinPCG: Dual Thread Redundancy with forward Recovery for Preconditioned Conjugate Gradient Methods

Dichev

Nikolopoulos

2016

2016 IEEE International Conference on Cluster Computing (CLUSTER)

Self Cite

View full text Add to dashboard Cite

Section: B Detection and Correctionsupporting

confidence: 59%

TwinPCG: Dual Thread Redundancy with forward Recovery for Preconditioned Conjugate Gradient Methods

Dichev

Nikolopoulos

2016

2016 IEEE International Conference on Cluster Computing (CLUSTER)

Self Cite

View full text Add to dashboard Cite

“…• If the inequality in D1 holds (always for both threads), either no faults have occurred, or faults have occurred that we consider insignificant. In other words, our use of residual norm in D1 allows us to use 1 as a significance filter for bit flips which may very often be insignificant [15], [16]. This also has the implication that the two threads may diverge up to a certain point.…”

Section: B Detection and Correctionmentioning

confidence: 99%

TwinPCG: Dual Thread Redundancy with Forward Recovery for Preconditioned Conjugate Gradient Methods

Dichev¹,

Nikolopoulos

2016

2016 IEEE International Conference on Cluster Computing (CLUSTER)

Self Cite

View full text Add to dashboard Cite

Even though iterative solvers like the Conjugate Gradients method (CG) have been studied for over fifty years, fault tolerance for such solvers has seen much attention in recent years. For iterative solvers, two major reliable strategies of recovery exist: checkpoint-restart for backward recovery, or some type of redundancy technique for forward recovery. Important redundancy techniques like ABFT techniques for sparse matrixvector products (SpMxV) have recently been proposed, which increase the resilience of CG methods. These techniques offer limited recovery options, and introduce a tolerable overhead. In this work, we study a more powerful resilience concept, which is redundant multithreading. It offers more generic and stronger recovery guarantees, including any soft faults in CG iterations (among others covering ABFT SpMxV), but also requires more resources. We carefully study this redundancy/efficiency conflict. We propose a fault tolerant CG method, called TwinCG, which introduces minimal wallclock time overhead, and significant advantages in detection and correction strategies. Our method uses Dual Modular Redundancy instead of the more expensive Triple Modular Redundancy; still, it retains the TMR advantages of fault correction. We describe, implement, and benchmark our iterative solver, and compare it in terms of efficiency and fault tolerance capabilities to state-of-the-art techniques. We find that before parallelization, TwinCG introduces around 5-6% runtime overhead compared to standard CG, and after parallelization efficiently uses BLAS. In the presence of faults, it reliably performs forward recovery for a range of problems, outperforming SpMxV ABFT solutions.

show abstract

“…Contrary to approximation techniques via loop perforation implemented in the compiler [13], our programming model uses domain expertise available from the programmer, which we demonstrate to be necessary for effective approximation in at least one application domain. On the other hand, our programming model remains general-purpose, contrary to applicationspecific approximate code generators such as SAGE [10], Paraprox [9] ApproxIt [20] and related work on iterative solvers [4] and video codecs [12]. Compared to other approximate language frameworks such as EnerJ [11], our programming model offers additional features including task-parallel execution and energy-constrained runtime optimization of output quality.…”

Section: Related Workmentioning

confidence: 99%

A significance-driven programming framework for energy-constrained approximate computing

Vassiliadis

Chalios

Parasyris

et al. 2015

Proceedings of the 12th ACM International Conference on Computing Frontiers

Self Cite

View full text Add to dashboard Cite

Approximate execution is a viable technique for energy-constrained environments, provided that applications have the mechanisms to produce outputs of the highest possible quality within the given energy budget.We introduce a framework for energy-constrained execution with controlled and graceful quality loss. A simple programming model allows users to express the relative importance of computations for the quality of the end result, as well as minimum quality requirements. The significance-aware runtime system uses an application-specific analytical energy model to identify the degree of concurrency and approximation that maximizes quality while meeting user-specified energy constraints.Evaluation on a dual-socket 8-core server shows that the proposed framework predicts the optimal configuration with high accuracy, enabling energy-constrained executions that result in significantly higher quality compared to loop perforation, a compiler approximation technique.

show abstract

On the potential of significance-driven execution for energy-aware HPC

Cited by 10 publications

References 19 publications

TwinPCG: Dual Thread Redundancy with forward Recovery for Preconditioned Conjugate Gradient Methods

TwinPCG: Dual Thread Redundancy with forward Recovery for Preconditioned Conjugate Gradient Methods

TwinPCG: Dual Thread Redundancy with Forward Recovery for Preconditioned Conjugate Gradient Methods

A significance-driven programming framework for energy-constrained approximate computing

Contact Info

Product

Resources

About