Fault-Tolerant Routing for Exascale Supercomputer: The BXI Routing Architecture

Vigneras, Pierre; Quintin, Jean-Noël

doi:10.1109/cluster.2015.135

Cited by 4 publications

(3 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This contrasts with common techniques based on simulation or experimentation which do not link observations of contention with a corresponding explanation. This simple technique of estimation of contention is new; it is also used in concert with the architecture described by Vigneras & Quintin [8] with the goal of automating computation of that metric for potential integration into the fabric management's decision making. Analysis in terms of this metric is sufficient to prove and explain drawbacks and benefits of algorithms, but a simulation-based analysis would complement this work to give tangible results for real-life applications.…”

Section: A Static Congestion Metricmentioning

confidence: 99%

“…No perfect agnostic algorithm exists, and designing a good routing algorithm usually requires paying attention to the topology and communication patterns which will take place. As detailed by Vigneras & Quintin [8], we can consider that the topology of an HPC cluster never changes, so algorithms are usually designed for a given topology class (i.e. topology-aware algorithms).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Node-Type-Based Load-Balancing Routing for Parallel Generalized Fat-Trees

Gliksberg¹,

Quintin²,

García³

2018

2018 IEEE 4th International Workshop on High-Performance Interconnection Networks in the Exascale and Big-Data Era (HiPINEB)

Self Cite

View full text Add to dashboard Cite

High-Performance Computing (HPC) clusters are made up of a variety of node types (usually compute, I/O, service, and GPGPU nodes) and applications don't use nodes of a different type the same way. Resulting communication patterns reflect organization of groups of nodes, and current optimal routing algorithms for all-to-all patterns will not always maximize performance for group-specific communications. Since application communication patterns are rarely available beforehand, we choose to rely on node types as a good guess for node usage. We provide a description of node type heterogeneity and analyse performance degradation caused by unlucky repartition of nodes of the same type. We provide an extension to routing algorithms for Parallel Generalized Fat-Tree topologies (PGFTs) which balances load amongst groups of nodes of the same type. We show how it removes these performance issues by comparing results in a variety of situations against corresponding classical algorithms.

show abstract

Section: A Static Congestion Metricmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Node-Type-Based Load-Balancing Routing for Parallel Generalized Fat-Trees

Gliksberg¹,

Quintin²,

García³

2018

2018 IEEE 4th International Workshop on High-Performance Interconnection Networks in the Exascale and Big-Data Era (HiPINEB)

Self Cite

View full text Add to dashboard Cite

show abstract

“…PQFT [4] is similar, though it requires a complete list of faults. The combination of Dmodk + Ftrnd diff [11] available in BXI FM [12] is applied in an offline/online manner (with an iterative list of topology changes and an up-to-date view of the topology), the goal being fast reaction to faults with minimal routing changes. Fabriscale [13] also provides fast centralized re-routing of fat-trees, by precomputing alternative routes.…”

Section: Introductionmentioning

confidence: 99%

High-Quality Fault-Resiliency in Fat-Tree Networks (Extended Abstract)

Gliksberg¹,

Capra²,

Louvet³

et al. 2019

2019 IEEE Symposium on High-Performance Interconnects (HOTI)

View full text Add to dashboard Cite

Coupling regular topologies with optimized routing algorithms is key in pushing the performance of interconnection networks of HPC systems. In this paper we present Dmodc, a fast deterministic routing algorithm for Parallel Generalized Fat-Trees (PGFTs) which minimizes congestion risk even under massive topology degradation caused by equipment failure. It applies a modulo-based computation of forwarding tables among switches closer to the destination, using only knowledge of subtrees for pre-modulo division. Dmodc allows complete rerouting of topologies with tens of thousands of nodes in less than a second, which greatly helps centralized fabric management react to faults with high-quality routing tables and no impact to running applications in current and future very large-scale HPC clusters. We compare Dmodc against routing algorithms available in the InfiniBand control software (OpenSM) first for routing execution time to show feasibility at scale, and then for congestion risk under degradation to demonstrate robustness. The latter comparison is done using static analysis of routing tables under random permutation (RP), shift permutation (SP) and all-to-all (A2A) traffic patterns. Results for Dmodc show A2A and RP congestion risks similar under heavy degradation as the most stable algorithms compared, and near-optimal SP congestion risk up to 1% of random degradation.

show abstract

Transitively Deadlock-Free Routing Algorithms

Quintin

Vigneras

2016

2016 2nd IEEE International Workshop on High-Performance Interconnection Networks in the Exascale and Big-Data Era (HiPINEB)

View full text Add to dashboard Cite

Fault-Tolerant Routing for Exascale Supercomputer: The BXI Routing Architecture

Cited by 4 publications

References 32 publications

Node-Type-Based Load-Balancing Routing for Parallel Generalized Fat-Trees

Node-Type-Based Load-Balancing Routing for Parallel Generalized Fat-Trees

High-Quality Fault-Resiliency in Fat-Tree Networks (Extended Abstract)

Transitively Deadlock-Free Routing Algorithms

Contact Info

Product

Resources

About