Aug 2026· Asia-Pacific Workshop on Networking· pp. 122-129· 0 citations· 36 references
Computer Science
TL;DR
An architecture that decouples complex policy enforcement from high-speed packet forwarding to support VPC semantics on back-end NICs and enable front-end/back-end integration is proposed, suggesting that commodity hardware can support both high-throughput AI training and flexible VPC features.
Abstract
While commodity NIC-based back-end AI networks offer ultra-high intra-cluster bandwidth for distributed training, their limited programmability and on-chip resources hinder the implementation of advanced VPC features such as fine-grained isolation and stateful security policies. Furthermore, access to resources within the VPC needs to be routed through the front-end DPU, which is shared by the scale-up domain. The mismatch between the front-end DPU’s bandwidth and the back-end requirements causes GPU underutilization when intensive VPC communication is required for content recommendation, AIGC, and federated learning workloads. We propose an architecture that decouples complex policy enforcement from high-speed packet forwarding to support VPC semantics on back-end NICs and enable front-end/back-end integration. Evaluations show near-full GPU utilization in our analytical model and 71 μ s P999 extra latency of the first packet, suggesting that commodity hardware can support both high-throughput AI training and flexible VPC features.
The experience in designing, deploying, and operating Pegasus, a data center network tailored for the AI cloud, along with operational lessons learned from its deployment are shared.
Xianneng Zou, Yadong Liu, Yiran Zhang et al.· Proceedings of the ACM SIGCO...· 0 citations
The deployment of real-time artificial intelligence inference services on shared cloud infrastructure poses significant challenges due to resource contention, latency variability, and tail-latency amplification. While cloud platforms offer scalability and flexibility, conventional accelerator sharing mechanisms often fail to provide the determinism required by latency-sensitive inference workloads. This paper presents a standards-based, multi-tenant cloud inference framework that integrates OpenStack orchestration with Single Root I/O Virtualization (SR-IOV)-enabled graphics processing unit (GPU) partitioning to achieve predictable and isolated real-time inference execution. In the proposed architecture, each tenant is assigned an exclusive GPU virtual function, enabling hardware-level isolation while remaining fully compatible with native OpenStack scheduling and resource management mechanisms. A comprehensive experimental evaluation is conducted on a private OpenStack cloud to assess inference latency distribution, tail behavior, scalability, robustness to background network and control-plane activity, and throughput-latency trade-offs. Experimental results show that median inference latency remains stable across single-tenant and multi-tenant configurations, while P95 and P99 tail latencies exhibit no measurable amplification under concurrent execution. The system scales linearly with the number of available GPU virtual functions, maintaining consistent latency behavior until hardware capacity is reached. Additional experiments demonstrate that background network traffic and control-plane operations introduce negligible impact on inference latency. Throughput analysis reveals a well-defined saturation knee, enabling clear identification of safe operating regions for real-time inference services. By leveraging mature ICT standards and open-source cloud infrastructure, this work provides a reusable reference architecture for deploying latency-sensitive inference services in private and hybrid clouds. The results highlight the effectiveness of hardware-assisted accelerator isolation in balancing performance determinism, scalability, and operational simplicity, and offer practical guidance for future system design and standardization efforts.
Mauricio Rui, Bingfeng Shi, Ma Xin et al.· Journal of ICT Standardizati...· 0 citations
The rapid proliferation of Artificial Intelligence (AI) workloads, including Large Language Models (LLMs) and Generative AI (GAI), is fundamentally transforming Multi-access Edge Computing (MEC) infrastructures. AI-driven services deployed at the edge exhibit highly synchronized, bursty and bandwidth-intensive traffic patterns, contrasting with traditional MEC applications that generate moderate, stable flows. This evolution stresses intra-MEC network fabrics, leading to potential congestion, unfair queuing and underutilized computational resources when conventional architectures designed for elastic workloads are applied. In response, we propose an AI-aware intra-MEC data center network (DCN) architecture coupled with Wavelength Division Multiplexing (WDM)-based Medium Access Control (MAC) protocols. The proposed edge DCN enables fine-grained traffic prioritization, ensuring latencysensitive AI workloads maintain sub-us latencies while efficiently accommodating medium and low-priority applications. Through simulation-driven evaluation under realistic AI escalation scenarios, we demonstrate over 90% bandwidth utilization with near-zero packet loss, deterministic latency guarantees across heterogeneous workloads, and robust performance even under extreme AI traffic bursts. These results highlight the potential of AI-aware, intra-MEC network architectures to meet the demands of AI-centric edge environments. Ultimately, this work contributes to the realization of next-generation networks and the MEC 2.0 vision, where intelligent, high-performance edge infrastructures seamlessly support diverse, latency-sensitive workloads at scale.
G. Drainakis, P. Baziana, A. Bogris· International Mediterranean...· 0 citations
Researchers and practitioners working with large language models face a fragmented landscape: local models are free and private but hardware limits the model size and context windows a researcher can use; institutional HPC centers offer powerful GPU resources at no marginal cost and keep data within institutional boundaries, but operate behind firewalls and are designed for batch jobs rather than interactive use; commercial cloud APIs provide frontier-model quality on demand but impose significant cost and data retention policies unsuitable for sensitive research data. No existing system unifies all three. STREAM (Smart Tiered Routing Engine for AI Models) addresses this gap with four contributions: (1) a three-tier routing architecture combining local, HPC, and cloud inference with a local LLM-based complexity judge; (2) a dual-channel HPC streaming architecture that separates the Globus Compute control plane (authentication and job dispatch) from a WebSocket relay data plane (token delivery), enabling sub-second TTFT (0.54 s median, 21.1 × over batch mode’s 11.40 s) through institutional firewalls without VPN or firewall rule changes, with end-to-end AES-256-GCM encryption ensuring the relay operator cannot read token payloads; (3) tier-aware context summarization that prevents long conversations from forcing simple queries onto expensive tiers; and (4) an HPC-as-API proxy mode that exposes HPC inference as an OpenAI-compatible endpoint callable from any standard client with no HPC expertise, a deployment pattern made practical only by the sub-second TTFT of contribution (2). Llama 3.2 3B achieves 85.1% free-tier retention on a 1,200-query benchmark spanning ten domains. Measured TTFT: 0.26 s local, 0.54 s HPC (relay), 1.68 s cloud.
A. Nassar, Steve Mohr, L. Apanasevich et al.· Practice and Experience in A...· 0 citations
Compute and memory resources in cloud environments are strictly managed and isolated by the control plane; in contrast, network resources lack equivalent management and isolation mechanisms. This best-effort treatment of networking leads to significant challenges for modern AI workloads, which have diverse and bandwidth-intensive communication patterns. Without fine-grained network resource control, these workloads suffer from interference, unpredictable throughput, and suboptimal cluster utilization. To address these issues, this paper demonstrates how network bandwidth can be elevated to a first-class, schedulable, and enforceable resource within Kubernetes, the de facto standard for cloud-native orchestration. We introduce a new scheduling capability that models network interfaces as allocatable resources and regulates bandwidth sharing through the Dynamic Resource Allocation (DRA) framework, with enforcement implemented using the Hierarchical Token Bucket (HTB) mechanism. We evaluate the system using multitenant AI workloads derived from real-world communication characteristics with a simulation-based approach and validate the proposed enforcement strategy in a real cluster. Results show that the proposed two-level bandwidth allocation improves tenant performance predictability and satisfaction while maintaining packed cluster utilization.
Gleam is a novel and network-efficient framework for task-generic GPU sharing across local-area CUDA devices, with three key contributions that reduce bandwidth overhead in CUDA API remoting through automatic model weight caching, and mitigate accumulated latency from frequent API calls by asynchronous execution.
Zhihao Xu, Hao Zhong, Zeting Zhou et al.· 0 citations