In remote direct memory access (RDMA)-based remote procedure call (RPC) systems, using reliable connection (RC) transport mode with one-sided verbs might face challenges from excessive queue pairs (QPs) and per-client request regions when there is high client concurrency. This can cause cache thrashing on the RDMA network interface card (RNIC) and CPU and restrict server-side scalability. To address this issue, this paper proposesRdmax, a highly-scalable RC-based RPC system incorporating two innovations. First, it leverages QP multiplexity to serve multiple independent clients simultaneously with one RC QP, breaking the one-to-one connection constraint of RC. This also allows to post responses for different clients in one batch. Second, it uses in-network request dispatching to consolidate per-client request regions into a shared one, while still ensuring conflict-free client writes.Rdmaxrealizes these innovations with a programmable network, which manages the states of RC QPs and the request region and modifies relevant fields in RDMA packets accordingly. In a dummy RPC benchmark,Rdmaxachieves up to$10.4\times $higher throughput. For online transaction processing,Rdmaxoutperforms two state-of-the-arts based on client grouping and unreliable datagram schemes, improving throughput by up to 44.3% and 45.7%, while reducing 99% tail latency by up to 66.8% and 40.0%, respectively.

As network monitoring is crucial for ensuring the performance of network operations, one key challenge is how to optimize the tradeoff between its overheads and accuracy. This paper proposes a smart reporting mechanism, namely SmtRPTG, which continuously optimizes the scheme of network status collecting and reporting to properly balance the tradeoff. SmtRPTG estimates the distributions of status data of various types based on sampled results, and lets network elements make local decisions on whether and what type of status data should be reported based on the estimations. We formulate a probabilistic model with hidden variables, based on which an algorithm is designed to estimate data distributions for SmtRPTG. Extensive simulations verify the effectiveness of our proposal on balancing the tradeoff between overheads and accuracy of network monitoring.

With the rapid development of network function virtualization (NFV), there is an increasing trend of implementing virtual network functions (vNFs), especially the stateful ones, on high-performance programmable data plane (PDP) switches (e.g., the P4-based PDP switches based on Tofino ASICs), and forming stateful service function chains (SFCs) with them. However, the capability of PDP switches on supporting stateful SFCs is still restricted by the limited hardware resources in them. In this work, we study how to optimize the deployment of stateful SFCs in P4-based PDP switches and design the system of st-SFC, so as to not only utilize the hardware resources on switches efficiently but also minimize the overhead of interactions between control and data planes. We first consider the deployment of stateful SFCs on a single PDP switch. Specifically, we propose to abstract each stateful vNF as a state machine and design a stateful SFC building algorithm to merge the state machines of vNFs for reducing redundant resource usages, and for the vNFs whose operations involve interactions with the control plane, we develop a PktIn-Table to reduce the resource usage in PDP switches and the interaction latency. Then, we propose an SFC deployment algorithm that realizes stateful SFCs on PDP switches on demand, aiming to optimize the resource usages across all the switches in runtime. We prototype st-SFC with PDP switches based on Tofino ASICs and demonstrate its effectiveness experimentally.

Data plane programmable (PDP) switches are becoming increasingly popular for network function virtualization (NFV), for their programmability and high packet processing performance. However, the inherent limitations of PDP switches, such as limited memory space, make it challenging to implement certain types of network functions (NFs) (i.e., the stateful ones) on them. This paper proposes SFCache, which combines PDP switches and commodity servers to achieve self-adaptive SFC deployment. SFCache aims to exploit the high packet processing performance of PDP switches while supporting the flexible deployment of a wide range of SFCs (including the stateful ones) with servers. Specifically, SFCache can dynamically improve the packet processing performance of the SFCs that were deployed on servers by selectively caching SFC-level packet processing rules on PDP switches. We design a few key components to facilitate SFCache, including an NF-destructed P4 pipeline that allows customizing packet processing rules in a match-rewrite pattern, a runtime NF synthesis method that can transform a set of NF-level match-rewrite rules into an equivalent SFC-level rule, and a count-min selection strategy to choose the best synthesized rule for being cached in PDP switch pipeline. We prototype SFCache with a PDP switch based on Tofino ASIC and a server, and demonstrate the effectiveness of our proposal experimentally.

With momentum gained from programmable data plane (PDP), in-band network telemetry (INT) has been widely considered as a promising technique for realtime network monitoring. In this work, we leverage network coding (NC) to design CodedINT, for improving the visibility of INT in lossy networks. Specifically, we propose to encode the telemetry data in multiple packets with NC and distribute the encoded data over a group of packets. Then, among the group of packets, if we can receive enough ones that satisfy the decoding condition of NC, the whole original telemetry data can be recovered. We first explain the design of CodedINT to elaborate on its operation principle, packet format, and system implementation. Then, we implement and experimentally evaluate CodedINT in a real network testbed. Our experiments demonstrate that in a lossy network with packet loss rate at 80%, CodedINT ensures that 89.48% of the telemetry data carried by INT packets can be recovered successfully.