In-network aggregation (INA) has emerged as a key technology to alleviate communication bottlenecks in large-scale distributed training, but its performance is often hindered by suboptimal routing. Existing INA-aware routing algorithms suffer from certain limitations: they either lack a global, multi-job coordination mechanism, or operate on incomplete network models that ignore key hardware constraints such as switch processing capacity. These deficiencies lead to network congestion and inefficient resource utilization, ultimately undermining the full potential of INA. To address these challenges, we present INARouting, a novel framework that holistically solves the multi-job hierarchical aggregation routing problem. We propose TINA, a hierarchical aggregation protocol that supports multi-job in-network aggregation. To address different deployment scenarios, we develop two variants: INARouting-Opt that provides optimal solutions for moderate-scale networks, and INARouting-Relax, a fast and effective heuristic using LP-relaxation and a greedy score-based rounding algorithm for large-scale deployments. Through extensive experiments on various scales of Fat-Tree and Spine-Leaf topologies, we demonstrate that INARouting significantly outperforms state-of-the-art methods. INARouting- Opt achieves provably optimal solutions, reducing average job completion time by up to 56% compared to existing methods. Meanwhile, INARouting-Relax outperforms existing algorithms while being 5× faster in solving time, enabling efficient routing in large-scale, dynamic environments.

To train large AI models, clusters are constructed with abundant connectivity and bandwidth; but the commodity protocol ECMP and recent proposals fail to fully utilize the network bandwidth for AI traffic pattern. As model training jobs and AI clusters exhibit a predictable and periodic traffic pattern, so in this paper, we propose a MOdel training Routing System — MORS — for traffic routing in AI clusters. MORS defines temporal attributes to characterize the periodic traffic pattern of flows and network links, and temporal quality to quantify whether a path could deliver a flow quickly in the near future. MORS runs In-band Network Telemetry (INT) to collect temporal attributes of the network, and periodic analysis to extend the collected attributes in the time domain. Based on the time series of link utilization and latency, MORS computes the temporal quality of candidate paths. It enforces high-quality path selection while maintaining compatibility with commodity ECMP by manipulating the source UDP port to ensure the flow complies with the target path in the ECMP protocol. MORS is light-weight and readily deployable in the RDMA commodity cluster. Our prototype and experiments demonstrate that MORS achieves performance comparable to adaptive routing and delivers up to 14% and 50% better FCT than PLB and ECMP, respectively.

For distributed training (DT) based on the parameter servers (PS) architecture, the communication overhead is huge in the network for servers synchronizing parameters. In the PS architecture, the workers send gradients over the network to PS for aggregation. With the development of programmable switches, in-network aggregation (INA) is proposed to accelerate distributed training by utilizing the programmable switches in the network to implement gradients aggregation, not only at PS. However, the existing routing methods can not fully utilize the capability of INA, resulting in load imbalance and long communication time. This paper analyzes and models the routing problem in INA under the constraint of network resources. And we propose a routing algorithm named AggTree to solve this problem by searching the high-rate routing path. The result of simulations shows that AggTree can reduce communication time by 4.1%-37.9% for a single DT job and 12.7%-74.0% for multiple DT jobs compared with state-of-the-art solutions.