| Full text | |
| Author(s): |
Nunes, Diego Cardoso
;
Coelho, Bruno Loureiro
;
Parizotto, Ricardo
;
Schaeffer-Filho, Alberto Egon
Total Authors: 4
|
| Document type: | Journal article |
| Source: | INTERNATIONAL JOURNAL OF NETWORK MANAGEMENT; v. N/A, p. 18-pg., 2024-07-24. |
| Abstract | |
Achieving high-performance aggregation is essential to scaling data-parallel distributed machine learning (ML) training. Recent research in in-network computing has shown that offloading the aggregation to the network data plane can accelerate the aggregation process compared to traditional server-only approaches, reducing the propagation delay and consequently speeding up distributed training. However, the existing literature on in-network aggregation does not provide ways to deal with slower workers (called stragglers). The presence of stragglers can negatively impact distributed training, increasing the time it takes to complete. In this paper, we present Serene, an in-network aggregation system capable of circumventing the effects of stragglers. Serene coordinates the ML workers to cooperate with a programmable switch using a hybrid synchronization approach where approaches can be changed dynamically. The synchronization can change dynamically through a control plane API that translates high-level code into switch rules. Serene switch employs an efficient data structure for managing synchronization and a hot-swapping mechanism to consistently change from one synchronization strategy to another. We implemented and evaluated a prototype using BMv2 and a Proof-of-Concept in a Tofino ASIC. We ran experiments with realistic ML workloads, including a neural network trained for image classification. Our results show that Serene can speed up training by up to 40% in emulation scenarios by reducing drastically the cumulative waiting time compared to a synchronous baseline. This study proposes a mechanism to handle stragglers in data-parallel ML training by offloading the control of the SSP synchronization to the programmable data plane using a memory-efficient data structure. This work also introduces an API for dynamic adjustment of synchronization barriers. Tested by injecting stragglers in a realistic ML workload, on both an emulated environment and real hardware, the system improves job completion time and waiting time with limited impact on model accuracy. image (AU) | |
| FAPESP's process: | 23/00764-2 - RESIST: Resilient In-Network Computing |
| Grantee: | Alberto Egon Schaeffer Filho |
| Support Opportunities: | Regular Research Grants |
| FAPESP's process: | 20/05152-7 - PROFISSA: Programmable Future Internet for Secure Software Architectures |
| Grantee: | Lisandro Zambenedetti Granville |
| Support Opportunities: | Research Projects - Thematic Grants |