Aug 2026· Proceedings of the ACM SIGCOMM 2026 Conference· 1 citation· 67 references
TL;DR
XFir is the first hardware-accelerated new-flow setup system for cloud host servers that delivers high CPS throughput while preserving sufficient flexibility, and is cost-effective to deploy, requiring only a single DPU per host.
Abstract
In today's cloud networks, host servers widely deploy Data Processing Units (DPUs) as network accelerators under the "Sep-Path" paradigm. However, as server capabilities scale with increasing CPU cores and network bandwidth, the software slow path (executed on a DPU's CPU) has become a critical bottleneck for workloads with high new-flow rates. Meanwhile, new-flow setup logic on host servers must continuously evolve to meet diverse and changing customer demands, making flexibility a key requirement alongside performance. To address this gap, we present XFir, the first hardware-accelerated new-flow setup system for cloud host servers that delivers high CPS throughput while preserving sufficient flexibility. XFir leverages a next-generation DPU equipped with a Cloud Network co-Processor (CNP) to execute the host server's new-flow setup logic. XFir redesigns the host-server flow-setup datapath and table layout, optimizes LPM lookups, and introduces CPU-CNP collaboration mechanisms to further improve performance and reliability. Our evaluation shows that XFir achieves over 776K new-flow CPS on a single host server with 11.7μs slow-path latency. Compared to prior work (Fornax), XFir achieves 4.8x CPS and reduces latency by 69.2%. Moreover, XFir is cost-effective to deploy, requiring only a single DPU per host. Overall, XFir improves new-flow throughput while maintaining development flexibility at low financial cost.
The transition to Data Processing Unit (DPU)-centric architectures has become the de-facto standard in modern cloud networks, enabling infrastructure offload and improved host resource utilization. However, the fixed hardware limits of DPUs increasingly fail to keep pace with the rapid growth of host compute density and network-intensive workloads. As a result, when DPU resources are saturated, host compute capacity often remains underutilized due to insufficient network provisioning. This paper presents Spillway, a unified vSwitching fabric that decouples virtual switching capacity from static DPU hardware limits. Spillway introduces a DPU-host hybrid data plane that repurposes idle host CPU resources to process spillover traffic when the DPU becomes the bottleneck. Unlike traditional software switching, Spillway activates host-based forwarding selectively through a unified resource scheduler, incurring no overhead under normal operating conditions. To ensure production viability, Spillway enforces strict performance isolation across heterogeneous data planes. Experimental results from a large-scale deployment show that, under bursty traffic workloads, Spillway improves single-node connection processing capacity by up to 90% with negligible interference to co-located applications.
Xiaochong Jiang, Dian Fan, Yilong Lv et al.· Proceedings of the ACM SIGCO...· 0 citations
The new EMC+ proposal is an OS‐driven elasticity manager for container‐based environments that continuously estimates idle core cycles left by regular (inelastic) applications, and reallocates idle cores to elastic ones, even during short time intervals, and has minimal impact on the performance and QoS of colocated inelastic applications.
J. C. Saez, Carlos Bilbao, Manuel Prieto-Matías· Concurrency and Computation· 0 citations
Today, AI cloud is key to serving diverse users with AI services, where cloud networking forms the basis. In this paper, we share our experience in designing, deploying, and operating Pegasus, a data center network tailored for the AI cloud, along with operational lessons learned from its deployment. The key designs of Pegasus include: 1) Network virtualization: a DPU-RNIC decoupled collaborative hardware architecture to enable a single DPU to virtualize multiple RNICs while reducing the power consumption. We design two-level flow tables on both DPU and RNICs to support underlay-overlay IP address translation and ensure isolation. For DPU-RNIC communication, we introduce a per-RNIC communication state machine to reduce communication overhead. 2) Network transport: customized and transparent transport offloading in the RNIC for low-latency and high-throughput communication performance for various AI workloads. We carefully offload per-packet load balancing and credit-based congestion control in RNICs, optimizing reorder delay and eliminating the impacts of hardware jitter. Pegasus has been deployed in production for over two years, currently covering 8K GPUs and supporting a wide range of tenants' AI applications.
Xianneng Zou, Yadong Liu, Yiran Zhang et al.· Proceedings of the ACM SIGCO...· 0 citations
Stateful Network Functions (NFs) are widely deployed in cloud environments, where operations such as scaling, failover, and migration depend on NF state to ensure correct processing. Supporting these operations requires state transfer between NFs. The emergence of Multi Access Edge Computing (MEC) introduces state transfer scenarios such as state splitting and merging across NFs, driven by user mobility and distributed deployment. Existing techniques perform state transfer at NFs, which becomes a bottleneck, particularly when a NF transfers state to other NFs while continuing packet processing or competing for CPU resources. We present BoostState, a low-latency state transfer framework in MEC that leverages programmable data planes to accelerate state transfer while preserving conventional cloud-based NF deployments. BoostState leverages state partitioning, where NF state is partitioned into header and action state. It delegates handling header state to programmable data planes, reducing state transfer overhead at NFs. We implement BoostState using P4 language on a Netronome SmartNIC. Our results show that BoostState reduces state transfer latency compared to prior work, including up to 60% lower transmission latency.
Mahdi Attawna, Tung V. Doan, F. Fitzek et al.· Applied Networking Research...· 0 citations
Current NFVs assign packets to CPU cores at flow granularity, where each flow is pinned to a single CPU. This approach is efficient under most scenarios but has exposed limitations when handling elephant flows. These “heavy hitters” overwhelm single cores, creating bottlenecks that affect overall throughput and degrade service quality. As networks scale to higher-speed links and core-rich CPUs, these imbalances become more severe. In this paper, we propose ParaFlowO, an architecture that Parallelizes processing elephant Flows across multiple CPU cores while preserving in-Order delivery. ParaFlowO breaks elephant flows into flowlets and dynamically rotates them across multiple cores. It integrates a lightweight reordering mechanism to preserve packet order and controls parallelism to mitigate contention on shared state. Preliminary evaluations show that ParaFlowO offers a practical solution to mixed-grained parallelism in stateful middleboxes.
Changgang Zheng, Bowen Yang, Jin Ke et al.· Asia-Pacific Workshop on Net...· 1 citation
Applications requiring low-latency packet processing are challenging for today’s network and service management when resources are limited and must be shared. Containers are suitable, but achieving connectivity between containers exclusively in software is unsuitable for low-latency requirements. The impact of network latencies in containerized environments has received comparatively less attention, particularly in comparison to research on virtual machines. This paper analyzes throughput and network latencies in container-based networks on a single host featuring single-root input/output virtualization, Linux Containers, and commercial off-the-shelf hardware. We conduct measurements using a state-of-the-art measurement methodology to identify tail-latency behavior, achieving a resolution of $\mathrm {1.25~ {\mu }\text {s}}$ . We evaluate a single flow in a line topology with up to 64 containers and a complex topology with 38 flows and 12 container nodes. The experiments demonstrate that pinning interrupt request handlers to non-uniform memory access nodes increases throughput and decreases latencies. Furthermore, we identify data translation lookaside buffer misses, rescheduling interrupts, and soft interrupt floods as critical challenges causing spikes in latencies while isolation remains impossible. Our findings identify bottlenecks for real-time container applications. A comparison with VM measurements shows that containers can achieve latencies up to $\mathrm {60~ {\mu }\text {s}}$ lower. We support in this paper, network and service management in deciding on the underlying virtualization technology for packet-processing applications by providing recommendations accordingly.
Florian Wiedner, Alexander Daichendt, Jonas Andre et al.· IEEE Transactions on Network...· 0 citations