Reproduce “SCENIC: Stream Computation-Enhanced SmartNIC” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Benjamin Ramhorst, Maximilian Jakob Heer, Luhao Liu, Heejae Kim, Jonas Dann, Jin-Soo Kim, Gustavo Alonso
SCENIC, an open-source FPGA-based SmartNIC developed at ETH Zurich and Seoul National University, integrates 200G line-rate networking with deep customization capabilities and native datacenter integration, including direct access to GPUs and SSDs. This platform reduces SSD offload latency by 2-3x and accelerates hash-based data partitioning by 6.7x compared to CPU baselines, offering a flexible and high-performance solution for evolving datacenter workloads.
Key insights
- The NIC datapath can serve as a primary stream computation substrate, facilitating tight hardware/software co-design for both network infrastructure and applications.
- Open-sourcing the entire transport pipeline and enabling dynamic partial reconfiguration allows for full customization of network protocols and congestion control algorithms.
- Effective multi-tenant isolation and resource fairness can be achieved through on-device memory virtualization and system-wide arbitration for concurrent offloads.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.