iLoad

P8: P4 With Predictable Packet Processing Performance

“…Static performance of software switches [28] and FPGA [27] running P4 programs, as well as performance variation during runtime reconfiguration [47], are explored in different works. On a smaller scale, researches have been performed on evaluating the performance of P4 targets executing specific functionalities, such as encryption [48] (on NetFPGA and SmartNIC), in-network event processor [49]- [51] (on SmartNIC and Tofino), and stateless load-balancing [52], [53] (on NetFPGA and Tofino). Instead of taking a network function written in P4 as a whole, we break it down into atomic constructs, which can be applied to all targets and assembled back to form different network functions.…”

LAMP: Prompt Layer 7 Attack Mitigation with Programmable Data Planes

“…Distributed data centers naturally have heterogeneous servers in terms of geographical locations, renewable energy and volatile resources. In iLoad [15], servers determine their workload, e.g., machine learning tasks or web traffic, based on locations and resources. A programmable switch receives those workload requests directly from servers, and makes adjustments to packet forwarding but with little input from the controller.…”

“…Chuang et al in [7] take into consideration a task execution length and use an OpenFlow-based controller to monitor and reschedule the data center jobs. iLoad [15] is an in-network green load balancing solution; however, it still requires the control plane to analyze and update the data plane registers during switch operations. Xu et al in [29] design a green scheduler for data center flows with deadlines.…”