Citation:

Ratan Raj Anandeshi, 2026. Resilient Infrastructure Design for Supercomputing Clusters Using GPU-Centric Architectures  Volume 6 Issue 2: 278-282.

Abstract:

The adoption of graphics processing units (GPUs) as major computational accelerators has played a major role in the rapid development of high-performance computing (HPC). Graphics processing units (GPUs) have become central accelerators in modern high-performance computing because they provide massive parallelism, high throughput, and improved performance-per-watt for suitable data-parallel workloads. However, the growing dependence on GPU-based infrastructure introduces new resilience challenges such as fault tolerance, thermal instability, communication bottlenecks and system integration heterogeneity. These challenges are exacerbated by the scale and complexity of modern exascale systems. This review critically examines resilient infrastructure design in GPU-centric supercomputing clusters. Among the major themes are hardware-level fault reduction, system-level redundancy, software-defined fault mitigation and network-aware scheduling. Available literature shows a significant advance in the computational performance, but the resilience mechanisms are usually reactive as opposed to predictive. Machine learning has emerged as a promising direction to integrate failure prediction and adaptive resource allocation, but scalability and generalizability are still problematic. Here, critical weaknesses are found in the integrated resilience strategies, inter-layer optimization solutions, and universal benchmarking schemes for GPU-based environments. This review synthesizes existing methodologies and mentions the necessity of the holistic design principles, which comprise hardware, software, and network layers. The review emphasizes proactive resilience strategies that guarantee reliable performance of the system while minimizing system disruption.

References:

[1] Owens, J. D., Houston, M., Luebke, D., Green, S., Stone, J. E., & Phillips, J. C. (2008). GPU computing. Proceedings of the IEEE, 96(5), 879–899.

[2] Cappello, F., Geist, A., Gropp, W., Kale, L., Kramer, B., & Snir, M. (2009). Toward exascale resilience. International Journal of High Performance Computing Applications, 23(4), 374–388.

[3] Mittal, S., & Vetter, J. (2015). A survey of methods for analyzing and improving GPU energy efficiency. ACM Computing Surveys, 47(2), 19–45.

[4] Dongarra, J., Heroux, M., & Luszczek, P. (2013). High-performance computing: Challenges and opportunities. International Journal of High Performance Computing Applications, 27(1), 4–11.

[5] Nickolls, J., Buck, I., Garland, M., & Skadron, K. (2008). Scalable parallel programming with CUDA. ACM Queue, 6(2), 40–53.

[6] Rech, P., Snir, M., & Reed, D. (2014). Reliability challenges in GPU-based systems. IEEE Transactions on Dependable and Secure Computing, 11(5), 420–432.

[7] Moody, A., Bronevetsky, G., Mohror, K., & de Supinski, B. R. (2010). Design, Modeling, and Evaluation of a Scalable Multi-level Checkpointing System. Proceedings of SC 2010: International Conference for High Performance Computing, Networking, Storage and Analysis.

[8] Kim, J., Dally, W., Scott, S., & Abts, D. (2008). Technology-driven, highly-scalable dragonfly topology. ACM SIGARCH Computer Architecture News, 36(3), 77–88.

[9] Di, S., & Cappello, F. (2016). Adaptive algorithm for minimizing checkpoint overhead. IEEE Transactions on Parallel and Distributed Systems, 27(4), 964–977.

[10] Gupta, S., & Aiken, A. (2011). Energy-efficient GPU computing. IEEE Micro, 31(3), 40–50.

[11] Ferreira, K., Stearley, J., Laros, J., Oldfield, R., Pedretti, K., Brightwell, R., & Riesen, R. (2011). Evaluating the viability of process replication reliability for exascale systems. SC Conference, 1(1), 1–12.

[12] Zheng, Z., Lan, Z., & Liu, Z. (2012). Exploring failure prediction for HPC systems. IEEE Transactions on Parallel and Distributed Systems, 23(10), 1888–1897.

[13] Luo, Y., & Li, X. (2018). GPU cluster optimization for HPC workloads. Future Generation Computer Systems, 79(1), 1–10.

[14] Jain, N., Bhatele, A., Gamblin, T., & Kale, L. (2014). Predicting application resilience using machine learning. SC Conference, 1(1), 1–12.

[15] Li, D., & Chen, Y. (2017). Thermal-aware GPU scheduling. IEEE Transactions on Computers, 66(6), 1038–1051.

[16] Heroux, M., & Dongarra, J. (2019). Toward resilient exascale computing. Journal of Supercomputing, 75(10), 6991–7007.

[17] Gupta, A., & Kumar, V. (2016). Performance modelling of GPU clusters. IEEE Transactions on Parallel and Distributed Systems, 27(12), 3540–3553.

[18] Wang, K., & Chen, M. (2014). Dynamic voltage scaling for GPU clusters. IEEE Transactions on Computers, 63(5), 1231–1243.

[19] Zhao, J., & Zeng, J. (2020). Fault tolerance in heterogeneous computing systems. Journal of Parallel and Distributed Computing, 140(1), 1–12.

[20] Patel, P., & Shah, M. (2018). Energy-aware scheduling in GPU clusters. Future Generation Computer Systems, 86(1), 1–12.

Keywords:

GPU Architectures, High-Performance Computing, Resilience Engineering, Supercomputing Clusters, System Reliability, Fault Tolerance.