English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Even though iterative solvers like the Conjugate Gradients method (CG) havebeen studied for over fifty years, fault tolerance for such solvers has seenmuch attention in recent years. For iterative solvers, two major reliablestrategies of recovery exist: checkpoint-restart for backward recovery, or sometype of redundancy technique for forward recovery. Important redundancytechniques like ABFT techniques for sparse matrix-vector products (SpMxV) haverecently been proposed, which increase the resilience of CG methods. Thesetechniques offer limited recovery options, and introduce a tolerable overhead.In this work, we study a more powerful resilience concept, which is redundantmultithreading. 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/efficiencyconflict. We propose a fault tolerant CG method, called TwinCG, whichintroduces minimal wallclock time overhead, and significant advantages indetection and correction strategies. Our method uses Dual Modular Redundancyinstead of the more expensive Triple Modular Redundancy; still, it retains theTMR advantages of fault correction. We describe, implement, and benchmark ouriterative solver, and compare it in terms of efficiency and fault tolerancecapabilities to state-of-the-art techniques. We find that beforeparallelization, TwinCG introduces around 5-6% runtime overhead compared tostandard CG, and after parallelization efficiently uses BLAS. In the presenceof faults, it reliably performs forward recovery for a range of problems,outperforming SpMxV ABFT solutions.

TwinCG: Dual Thread Redundancy with Forward Recovery for Conjugate Gradient Methods