Xiangrui Yang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Xiangrui Yang has a total of 61 co-authors. Here are the top co-authors by number of shared publications:
- Huan Zhou
College of Computer Science and Technology, National University of Defense Technology, Changsha, China
4 shared publications - Wei Quan
National University of Defense Technology
3 shared publications - Wenfei Wu
Peking University, Beijing, China
3 shared publications - Yingwen Chen
College of Systems Engineering, National University of Defense Technology, Changsha, China
3 shared publications Jinfeng Huang
National Univ. of Defense Technology (China)
2 shared publications- Jinshu Su
National university of defense technology, Changsha, China
2 shared publications
Latest Publications
- Open Access
- 11 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 368-395
EPIC: Abstraction and Polymorphism of In-Network Collectives on Ethernet
- 11 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 368-395
Open AccessEPIC: Abstraction and Polymorphism of In-Network Collectives on Ethernet
- 25 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Networking
- Vol. 33 (6) , 3009-3023
FooDog: Empower TSN for Efficient Policing
Time-Sensitive Networking (TSN) is an emerging real-time Ethernet technology that provides deterministic communication for time-sensitive (TS) traffic. At its core, TSN utilizes Per-Stream Filtering and Policing (PSFP) gates to mitigate the disruption of unavoidable frame drift. However, as first identified in this work, the naive PSFP gate design results in heavy memory usage, which hinders normal switching functions. This work proposes an efficient PSFP gate design called FooDog. FooDog employs a two-stage structure and a dual-engine policing mechanism to realize memory-efficient, logic-compact, and fast policing while maintaining minimal latency and jitter for TS traffic. Results on FPGA prototypes show that FooDog consumes only hundreds of kilobits of memory, reducing on-chip memory overheads by more than 90% compared to the unoptimized PSFP gate design. Additionally, it maintains end-to-end latency in the microsecond range and jitter below 150 nanoseconds under abnormal traffic conditions, comparable to typical TSN performance without anomalies.
- 25 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Networking
- Vol. 33 (6) , 3009-3023
FooDog: Empower TSN for Efficient Policing
Time-Sensitive Networking (TSN) is an emerging real-time Ethernet technology that provides deterministic communication for time-sensitive (TS) traffic. At its core, TSN utilizes Per-Stream Filtering and Policing (PSFP) gates to mitigate the disruption of unavoidable frame drift. However, as first identified in this work, the naive PSFP gate design results in heavy memory usage, which hinders normal switching functions. This work proposes an efficient PSFP gate design called FooDog. FooDog employs a two-stage structure and a dual-engine policing mechanism to realize memory-efficient, logic-compact, and fast policing while maintaining minimal latency and jitter for TS traffic. Results on FPGA prototypes show that FooDog consumes only hundreds of kilobits of memory, reducing on-chip memory overheads by more than 90% compared to the unoptimized PSFP gate design. Additionally, it maintains end-to-end latency in the microsecond range and jitter below 150 nanoseconds under abnormal traffic conditions, comparable to typical TSN performance without anomalies.
- 24 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Services Computing
- Vol. 18 (5) , 2994-3007
Magneto: Load-Balanced Key-Value Service for Write-Intensive Workloads
High-performance key-value (KV) storage is critical for the cloud, providing KV services to various cloud applications. A key challenge for KV services is that workloads of cloud applications are often write-intensive and exhibit highly-skewed characteristics, which result in load imbalance among storage servers thus lowering system performance. To address this problem, prior works set up an in-switch write-back caching mechanism, which adopts a centralized controller to balance writes. However, the limited bandwidth between the switches and the controller is a severe bottleneck for achieving high performance. In this paper, we present Magneto, a novel key-value service architecture for the cloud. At the core of Magneto is a delayed-write mechanism in the switch data plane to absorb frequent write queries for hot items. It effectively balances the load under write-intensive workloads without involving the controller. Magneto also designs a reliability mechanism to ensure system reliability during switch state transitions and failures. We implement a prototype using an FPGA-integrated switch, which has high packet-processing performance and contains enough memory to provide both the in-switch cache and the write buffer. Extensive evaluation shows that Magneto can achieve 8.4x system throughput gains compared to baseline systems when handling a skewed workload consisting of 70% reads and 30% writes. Moreover, it can reduce the load on back-end servers up to 56% in total.
- 24 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Services Computing
- Vol. 18 (5) , 2994-3007
Magneto: Load-Balanced Key-Value Service for Write-Intensive Workloads
High-performance key-value (KV) storage is critical for the cloud, providing KV services to various cloud applications. A key challenge for KV services is that workloads of cloud applications are often write-intensive and exhibit highly-skewed characteristics, which result in load imbalance among storage servers thus lowering system performance. To address this problem, prior works set up an in-switch write-back caching mechanism, which adopts a centralized controller to balance writes. However, the limited bandwidth between the switches and the controller is a severe bottleneck for achieving high performance. In this paper, we present Magneto, a novel key-value service architecture for the cloud. At the core of Magneto is a delayed-write mechanism in the switch data plane to absorb frequent write queries for hot items. It effectively balances the load under write-intensive workloads without involving the controller. Magneto also designs a reliability mechanism to ensure system reliability during switch state transitions and failures. We implement a prototype using an FPGA-integrated switch, which has high packet-processing performance and contains enough memory to provide both the in-switch cache and the write buffer. Extensive evaluation shows that Magneto can achieve 8.4x system throughput gains compared to baseline systems when handling a skewed workload consisting of 70% reads and 30% writes. Moreover, it can reduce the load on back-end servers up to 56% in total.
- Open Access
- 12 August 2024
- Published by Association for Computing Machinery (ACM)
- p. 701-711
DeInfer: A GPU resource allocation algorithm with spatial sharing for near-deterministic inferring tasks
For the applications of artificial intelligence, training models with GPUs are widely noted, while inferring requirements are somehow neglected. In some scenario, it is quite important to finish the deep learning inference (DLI) task and get the response in time, e.g., anomaly detection in AIOps or QoE (Quality of Experience) assurance for customers. However, the challenges of GPU inferring are as follows: 1) the interference among inferring tasks on the shared GPU are not well studied and even not considered, which may cause a surge in the inference latency due to the hardware contention; 2) the deadline miss rate caused by the arrival rate is not clearly considered, which often exhibits significant fluctuations in real-world cases. Therefore, the interference among tasks and arrival rates of tasks should be well designed to decrease the deadline miss rate, when sharing GPU resources. To tackle the issue, we propose the algorithm, DeInfer, through following manners: 1) we identify the key factors that lead to interference, conduct a systematic study and develop a highly accurate interference prediction algorithm based on the random forest algorithm, achieving a four times improvement compared to the state-of-the-art interference prediction algorithms; 2) we utilize the queue theory to model the randomness of the arrival process and put forward a GPU resource allocation algorithm, which reduces the deadline miss rate by an average of over 30%.
- 12 August 2024
- Published by Association for Computing Machinery (ACM)
- p. 701-711
Open AccessDeInfer: A GPU resource allocation algorithm with spatial sharing for near-deterministic inferring tasks
For the applications of artificial intelligence, training models with GPUs are widely noted, while inferring requirements are somehow neglected. In some scenario, it is quite important to finish the deep learning inference (DLI) task and get the response in time, e.g., anomaly detection in AIOps or QoE (Quality of Experience) assurance for customers. However, the challenges of GPU inferring are as follows: 1) the interference among inferring tasks on the shared GPU are not well studied and even not considered, which may cause a surge in the inference latency due to the hardware contention; 2) the deadline miss rate caused by the arrival rate is not clearly considered, which often exhibits significant fluctuations in real-world cases. Therefore, the interference among tasks and arrival rates of tasks should be well designed to decrease the deadline miss rate, when sharing GPU resources. To tackle the issue, we propose the algorithm, DeInfer, through following manners: 1) we identify the key factors that lead to interference, conduct a systematic study and develop a highly accurate interference prediction algorithm based on the random forest algorithm, achieving a four times improvement compared to the state-of-the-art interference prediction algorithms; 2) we utilize the queue theory to model the randomness of the arrival process and put forward a GPU resource allocation algorithm, which reduces the deadline miss rate by an average of over 30%.
- Open Access
- 3 August 2024
- Published by Association for Computing Machinery (ACM)
- p. 196-197
A Geohash-based Naming Scheme for NDN Optimization
- 3 August 2024
- Published by Association for Computing Machinery (ACM)
- p. 196-197
Open AccessA Geohash-based Naming Scheme for NDN Optimization
- Open Access
- 3 August 2024
- Published by Association for Computing Machinery (ACM)
- p. 211-212
WeMu: A design of wireless network emulator
- 3 August 2024
- Published by Association for Computing Machinery (ACM)
- p. 211-212
Open AccessWeMu: A design of wireless network emulator
- Open Access
- 9 October 2023
- Published by Springer Nature in Journal of Cloud Computing
- Vol. 12 (1) , 1-13
Making programmable packet scheduling time-sensitive with a FIFO queue
Time-Sensitive Networking (TSN) is an emerging technology for real-time and non-real-time hybrid networked systems. TSN is standardized by IEEE 802.1 TSN Task Group and is becoming widely used in various scenarios including the cloud network. However, existing programmable packet schedulers such as PIFO, PIEO, and AIFO in programmable switches either lack the ability to express most scheduling algorithms in TSN or introduce intolerable on-chip memory overhead (e.g., strict-priority queues). This makes programmable switches and NICs incapable of providing deterministic forwarding. In this paper, we present AIAO (Admission-In-Admission-Out), a new set of programmable scheduling primitives using just a single FIFO to support typical TSN scheduling algorithms, as well as other popular work-conserving algorithms. AIAO is inspired by AIFO but improves it with a group of high-speed packet ingress/egress admission control triggered by high-precise and globally synchronized time, thus being able to support time-sensitive scheduling. We implement AIAO and evaluate it with FPGA-based TSN switches. The preliminary results show that AIAO guarantees correctness for a typical TSN scheduling algorithm with minimal logic and memory overhead.
- 9 October 2023
- Published by Springer Nature in Journal of Cloud Computing
- Vol. 12 (1) , 1-13
Open AccessMaking programmable packet scheduling time-sensitive with a FIFO queue
Time-Sensitive Networking (TSN) is an emerging technology for real-time and non-real-time hybrid networked systems. TSN is standardized by IEEE 802.1 TSN Task Group and is becoming widely used in various scenarios including the cloud network. However, existing programmable packet schedulers such as PIFO, PIEO, and AIFO in programmable switches either lack the ability to express most scheduling algorithms in TSN or introduce intolerable on-chip memory overhead (e.g., strict-priority queues). This makes programmable switches and NICs incapable of providing deterministic forwarding. In this paper, we present AIAO (Admission-In-Admission-Out), a new set of programmable scheduling primitives using just a single FIFO to support typical TSN scheduling algorithms, as well as other popular work-conserving algorithms. AIAO is inspired by AIFO but improves it with a group of high-speed packet ingress/egress admission control triggered by high-precise and globally synchronized time, thus being able to support time-sensitive scheduling. We implement AIAO and evaluate it with FPGA-based TSN switches. The preliminary results show that AIAO guarantees correctness for a typical TSN scheduling algorithm with minimal logic and memory overhead.
- Open Access
- 15 September 2022
- Published by MDPI AG in Electronics
- Vol. 11 (18) , 2920
A Programmable SRv6 Processor for SFC
A reconfigurable match table (RMT) is a programmable pipeline architecture for packet processing. In order to enable the programmable data plane to support segment routing IPv6 (SRv6) and other network protocols, this paper extends the deparser based on RMT. The paper uses the extended deparser and two RMT pipelines to build a protocol-independent network slice programmable data plane model named as the programmable SRv6 processor. We design it primarily for segment identifier (SID) processing of SRv6. We have proved that it can support SRv6, multiple semantics for SIDs, micro segment ID, multi-protocol label switching, and a service function chain (SFC). This architecture has broad application prospects. Experimental results on an FPGA showed that the extended deparser could achieve 100 Gbps throughput for 512B packets with few resources.
- 15 September 2022
- Published by MDPI AG in Electronics
- Vol. 11 (18) , 2920
Open AccessA Programmable SRv6 Processor for SFC
A reconfigurable match table (RMT) is a programmable pipeline architecture for packet processing. In order to enable the programmable data plane to support segment routing IPv6 (SRv6) and other network protocols, this paper extends the deparser based on RMT. The paper uses the extended deparser and two RMT pipelines to build a protocol-independent network slice programmable data plane model named as the programmable SRv6 processor. We design it primarily for segment identifier (SID) processing of SRv6. We have proved that it can support SRv6, multiple semantics for SIDs, micro segment ID, multi-protocol label switching, and a service function chain (SFC). This architecture has broad application prospects. Experimental results on an FPGA showed that the extended deparser could achieve 100 Gbps throughput for 512B packets with few resources.
- 7 March 2022
- Published by SPIE-Intl Soc Optical Eng
- p. 74
Programmable NIC-based task offload framework
SPIE Digital Library Proceedings
- 7 March 2022
- Published by SPIE-Intl Soc Optical Eng
- p. 74
Programmable NIC-based task offload framework
SPIE Digital Library Proceedings
- Open Access
- 31 December 2019
- Published by MDPI AG in Electronics
- Vol. 9 (1) , 59
DrawerPipe: A Reconfigurable Pipeline for Network Processing on FPGA-Based SmartNIC
In the public cloud, FPGA-based SmartNICs are widely deployed to accelerate network functions (NFs) for datacenter operators. We argue that with the trend of network as a service (NaaS) in the cloud is also meaningful to accelerate tenant NFs to meet performance requirements. However, in pursuit of high performance, existing work such as AccelNet is carefully designed to accelerate specific NFs for datacenter providers, which sacrifices the flexibility of rapidly deploying new NFs. For most tenants with limited hardware design ability, it is time-consuming to develop NFs from scratch due to the lack of a rapidly reconfigurable framework. In this paper, we present a reconfigurable network processing pipeline, i.e., DrawerPipe, which abstracts packet processing into multiple “drawers” connected by the same interface. NF developers can easily share existing modules with other NFs and simply load core application logic in the appropriate “drawer” to implement new NFs. Furthermore, we propose a programmable module indexing mechanism, namely PMI, which can connect “drawers” in any logical order, to perform distinct NFs for different tenants or flows. Finally, we implemented several highly reusable modules for low-level packet processing, and extended four example NFs (firewall, stateful firewall, load balancer, IDS) based on DrawerPipe. Our evaluation shows that DrawerPipe can easily offload customized packet processing to FPGA with high performance up to 100 Mpps and ultra-low latency (<10 µs). Moreover, DrawerPipe enables modular development of NFs, which is suitable for rapid deployment of NFs. Compared with individual NF development, DrawerPipe reduces the line of code (LoC) of the four NFs above by 68%.
- 31 December 2019
- Published by MDPI AG in Electronics
- Vol. 9 (1) , 59
Open AccessDrawerPipe: A Reconfigurable Pipeline for Network Processing on FPGA-Based SmartNIC
In the public cloud, FPGA-based SmartNICs are widely deployed to accelerate network functions (NFs) for datacenter operators. We argue that with the trend of network as a service (NaaS) in the cloud is also meaningful to accelerate tenant NFs to meet performance requirements. However, in pursuit of high performance, existing work such as AccelNet is carefully designed to accelerate specific NFs for datacenter providers, which sacrifices the flexibility of rapidly deploying new NFs. For most tenants with limited hardware design ability, it is time-consuming to develop NFs from scratch due to the lack of a rapidly reconfigurable framework. In this paper, we present a reconfigurable network processing pipeline, i.e., DrawerPipe, which abstracts packet processing into multiple “drawers” connected by the same interface. NF developers can easily share existing modules with other NFs and simply load core application logic in the appropriate “drawer” to implement new NFs. Furthermore, we propose a programmable module indexing mechanism, namely PMI, which can connect “drawers” in any logical order, to perform distinct NFs for different tenants or flows. Finally, we implemented several highly reusable modules for low-level packet processing, and extended four example NFs (firewall, stateful firewall, load balancer, IDS) based on DrawerPipe. Our evaluation shows that DrawerPipe can easily offload customized packet processing to FPGA with high performance up to 100 Mpps and ultra-low latency (<10 µs). Moreover, DrawerPipe enables modular development of NFs, which is suitable for rapid deployment of NFs. Compared with individual NF development, DrawerPipe reduces the line of code (LoC) of the four NFs above by 68%.