Reducing P4 language's voluminosity using higher-level constructs

Alcoz, Albert Gran; Busse-Grawitz, Coralie; Marty, Éric; Vanbever, Laurent

doi:10.1145/3565475.3569078

Cited by 4 publications

(4 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, implementing these complex queue disciplines at linerate (100Gbps) is impossible for current commercial switches due to resource constraints. Existing approximating fair queue scheduling methods (e.g., AFQ [87], PCQ [88], SP-PIFO [2], AIFO [107], and Cebinae [106]), usually require new hardware features such as programmable switching ASICs and more queuing resources, which are less supported in production environments. Current commercial switches usually set a static queue number and share buffers across ports, resulting in queue sharing between flows.…”

Section: Discussionmentioning

confidence: 99%

“…It is difficult for them to probe network performance metrics (e.g., network latency [63] or bandwidth bottleneck [65], [30]), because RoCEv2 traffic and other TCP/UDP traffic are assigned in different queues with different schedule policies in general. (2) The attackers are also assumed to have prior knowledge of a specific set of network flows that are expected to be cut off (we refer to them as target flows). This can be specified in advance by the Attack-as-a-Service buyers [73], [85].…”

Section: A Threat Modelmentioning

confidence: 99%

See 1 more Smart Citation

LoRDMA: A New Low-Rate DoS Attack in RDMA Networks

Wang,

Zhang,

et al. 2024

Proceedings 2024 Network and Distributed System Security Symposium

View full text Add to dashboard Cite

RDMA is being widely used from private data center applications to multi-tenant clouds, which makes RDMA security gain tremendous attention. However, existing RDMA security studies mainly focus on the security of RDMA systems, and the security of the coupled traffic control mechanisms (represented by PFC and DCQCN) in RDMA networks is largely overlooked. In this paper, through extensive experiments and analysis, we demonstrate that concurrent short-duration bursts can cause drastic performance loss on flows across multiple hops via the interaction between PFC and DCQCN. And we also summarize the vulnerabilities between the performance loss and the burst peak rate, as well as the duration. Based on these vulnerabilities, we propose the LoRDMA attack, a lowrate DoS attack against RDMA traffic control mechanisms. By monitoring RTT as the feedback signal, LoRDMA can adaptively 1) coordinate the bots to different target switch ports to cover more victim flows efficiently; 2) schedule the burst parameters to cause significant performance loss efficiently. We conduct and evaluate the LoRDMA attack at both ns-3 simulations and a cloud RDMA cluster. The results show that compared to existing attacks, the LoRDMA attack achieves higher victim flow coverage and performance loss with much lower attack traffic and detectability. And the communication performance of typical distributed machine learning training applications (NCCL Tests) in the cloud RDMA cluster can be degraded from 18.23% to 56.12% under the LoRDMA attack.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: A Threat Modelmentioning

confidence: 99%

LoRDMA: A New Low-Rate DoS Attack in RDMA Networks

Wang,

Zhang,

et al. 2024

Proceedings 2024 Network and Distributed System Security Symposium

View full text Add to dashboard Cite

show abstract

“…Currently, these innovative programmable packet schedulers are only prototypes and are not available yet in commodity switches. The authors in [28] provided an approximate PIFO scheduler (SP-PIFO) by using native strict-priority queues in a P4 switch. Although SP-PIFO is implemented in P4 hardware switches, its purpose is to approximate the PIFO scheduler rather than provide fair bandwidth allocation among competing TCP and UDP flows.…”

Section: Related Workmentioning

confidence: 99%

Providing Near Per-Flow Scheduling in Commodity Switches Without Per-Flow Queues

et al. 2023

View full text Add to dashboard Cite

Network quality of service (QoS) is essential for network applications. For many applications, getting a fair share of available bandwidth for their flows can prevent them from being blocked by other flows that do not respond to congestion. Providing per-flow scheduling in each output port of a commodity switch can isolate the flows that compete for the bandwidth of a bottleneck link. Although per-flow scheduling can maintain fair shares among competing flows, due to the high implementation costs of providing per-flow queues in commodity switches, this capability is rarely provided in commodity switches on the market. To address this need, we design and implement a near-per-flow scheduling scheme named Near Per-flow Scheduling (NPFS) in P4 programmable hardware switches and evaluate its performance. NPFS provides near-per-flow scheduling effectiveness in commodity switches that do not have per-flow queues in their output ports. NPFS utilizes the priority queues provided in most commodity switches and dynamically assigns competing flows to these queues based on their protocol types and current sending rates. Experimental results show that, when the number of competing flows is less than three times the number of queues, NPFS guarantees that the achieved bandwidths of these flows only deviate from their ideal fair shares by 5%.INDEX TERMS Near per-flow scheduling (NPFS), per-flow scheduling, programmable switches, programming protocol-independent packet processors.

show abstract

“…Legacy AQM algorithms have been adapted and implemented on P4 switches with certain limitations [65,66,67,68,69,70]. P4 also enabled the fast development of custom AQM algorithms such as P4-SRPT [71], P4-ABC [72], SP-FIO [73], FDPA [74], P4QoS [75] and QoSTCP [76]. The above schemes benefit from the advantage of accurate measurements on the data plane, which allows for more timely control actions on the network.…”

Section: Rate Control Supported By Programmable Data Planesmentioning

confidence: 99%

Explicit feedback for congestion control in Science DMZ cyberinfrastructures based on programable data-plane switches.

Figueroa,

Pezoa

View full text Add to dashboard Cite

Diariamente, se genera una gran cantidad de datos científicos en diferentes áreas del conoci-miento. Los desafíos de investigación son cada vez más sofisticados y demandan una mayor cantidad de información para tomar decisiones, y además, las herramientas tecnológicas están cada vez más al alcance de los investigadores. En muchos casos, esta información debe ser recopilada y procesada, y transmitida entre diferentes centros de investigación ubicados a considerable distancia. Se recurre Internet y otras ciber infraestructuras dedicadas para lograr este objetivo. La transferencia de información científica entre sitios remotos implica desafíos en el rendimiento, la seguridad, la coexistencia y la asignación de recursos. Por lo tanto, ESnet propuso el concepto de Science DMZ que proporciona patrones de diseño para un entorno de red optimizado para intercambiar datos científicos. Debido a su naturaleza, los flujos de datos científicos masivos son orientados a la conexión, demandando un alto y constante tasa de transmisión, con baja latencia y variabilidad deseables, para alcanzar tiempos de transmisión razonables. Esto impone desafíos fundamentales al diseño de la red, especialmente en los mecanismos de control de congestión, estimación de los buffers y detección de anomalías, temas objeto de estudio de la presente tesis. Esta disertación aborda dos tecnologías disruptivas para responder a estos desafíos: modelos impulsados por datos y dispositivos programables en el plano de datos. Las soluciones fueron evaluadas empleando redes de producción y testbeds utilizando dispositivos reales de enrutamiento y procesamiento en el plano de datos. Los resultados mostraron que las soluciones desarrolladas podrían mejorar efectivamente el rendimiento de las redes académicas y adaptar efectivamente los patrones de diseño de Science DMZ a redes no dedicadas.

show abstract

Reducing P4 language's voluminosity using higher-level constructs

Cited by 4 publications

References 9 publications

LoRDMA: A New Low-Rate DoS Attack in RDMA Networks

LoRDMA: A New Low-Rate DoS Attack in RDMA Networks

Providing Near Per-Flow Scheduling in Commodity Switches Without Per-Flow Queues

Explicit feedback for congestion control in Science DMZ cyberinfrastructures based on programable data-plane switches.

Contact Info

Product

Resources

About