Multi-tenant GPU clusters are designed to concurrently run multiple distributed ML training workloads. However, frequent data transfers among GPUs via collective operations can slow down training, as collectives from different tenants compete for network bandwidth. Recent work (e.g., CASSINI) has considered collective coordination to prevent network contention, but primarily focused on static job-level optimizations at deployment time, oblivious to runtime network conditions and the specific traffic pattern of each workload. In this paper, we present Symphony, an application-layer solution that dynamically coordinates collective operations across tenants at runtime. Symphony integrates seamlessly with existing clusters with minimal modifications to the collective communication library and includes a lightweight online scheduling mechanism that requires no advance information about the workloads or their collectives. We evaluate Symphony using both a real GPU testbed implementation and trace-driven simulations. Specifically, using realistic ML workloads in our testbed, we observe improvements of up to 13.2% in average communication time and 9.6% in training time compared to state-of-the-art solutions.

Multi-dimensional networks have become the standard for interconnecting GPUs to support large-scale distributed machine learning (ML) training, driven by increasing computational and memory demands. However, the inherent bandwidth heterogeneity across network dimensions poses a significant challenge in efficiently scheduling collective operations, such as AllReduce, to minimize communication time. Existing hierarchical approaches suffer from inefficiencies due to static or suboptimal pipeline structures with uniform-sized chunks. We propose HiPIPE, a novel solution for multi-dimensional All-Reduce collective communication, incorporating three key optimizations: chunk scheduling across all dimensions, adaptive chunk sizing, and dynamic chunk number selection. These techniques collectively enhance communication efficiency and maximize bandwidth utilization. Particularly, we formulate chunk size optimization as a linear programming problem to derive the optimal chunk sizes, significantly reducing pipeline bubbles and improving overall performance. Extensive experiments using real-testbed emulation demonstrate that HiPIPE consistently outperforms state-of-the-art (SOTA) approaches across diverse network topologies and configurations, achieving an average communication time reduction of 28.85% while sustaining bandwidth utilization of up to 98.11%.

Network topology obfuscation (NTO) is generally considered as a promising proactive mechanism to mitigate traffic analysis attacks. The main challenge is to strike a balance among energy consumption, reliable routing, and security levels due to resource constraints in sensor nodes. Furthermore, software-defined wireless sensor networks (WSNs) are more vulnerable to traffic analysis attacks due to the uncovered pattern of control traffic between the controller and the nodes. In this article, a new energy-aware NTO mechanism is proposed, which maximizes the attack costs and is efficient and practical to be deployed. Specifically, first, a route obfuscation method is proposed by utilizing ranking-based route mutation, based on four different critical criteria: 1) route overlapping; 2) energy consumption; 3) link costs; and 4) node reliability. Then, a sink node obfuscation method is introduced by selecting several fake sink nodes that are indistinguishable from actual sink nodes, according to the$k$-anonymity model. As a result, the most suitable routes and sink nodes can be selected, and a highest obfuscation level can be reached without sacrificing energy efficiency. Finally, extensive simulation results demonstrate that the proposed methods can strongly mitigate traffic analysis attacks and achieve effective NTO for software-defined WSNs. In addition, the proposed methods can reduce the success rate of the attacks while achieving lower energy consumption and higher network lifetime.

Software-defined wireless sensor networks (SDWSNs) can be functionally affected by malicious sensor nodes that perform arbitrary actions, e.g., message dropping or flooding. The malicious nodes can degrade the availability of the network due to in-band communications and the inherent lack of secure channels in SDWSNs. In this article, we design a hierarchical trust management scheme for SDWSNs (namely, TSW) to detect potential threats inside SDWSNs while promoting node cooperation and supporting decision making in the forwarding process. TSW evaluates the trustworthiness of involved nodes and enables the detection of malicious behavior at various levels of the SDWSN architecture. We develop sensitive trust computational models to detect several malicious attacks. Furthermore, we propose separate trust scores and parameters for control and data traffic, respectively, to enhance the detection performance against attacks directed at the crucial traffic of the control plane. Furthermore, we develop an acknowledgment-based trust recording mechanism by exploiting some built-in SDN control messages. To ensure the resilience and honesty of the trust scores, a weighted averaging approach is adopted, and a reliability trust metric is defined. Through extensive analyses and numerical simulations, we demonstrate that TSW is efficient in detecting malicious nodes that launch several communications and trust management threats, such as black-hole, selective forwarding, denial of service, bad mouthing, and ON–OFF attacks.

In WSNs, packets are delivered through mostly static shortest paths to their destination. However, static packet delivery makes WSNs highly vulnerable to traffic analysis attacks due to open area deployment. Existing defence proposals fail to achieve a balance between the protection level and the resource constraints. In this paper, we present a proactive SDN-based Route Mutation (SRRM) scheme that enables changing the routes of the multiple flows in WSNs simultaneously to defend against passive and stealthy reconnaissance and sniffer attacks while preserving reliable and energy-aware routing. Multiple routes are ranked for packet flow based on node reliability, energy consumption, link cost, and route overlapping. Our extensive simulation results show that these techniques can effectively provide route obfuscation for software-defined WSNs.