This section lists the outcomes of the four technical work packages in RED-SEA – publications, public deliverables and articles. The architecture figure below shows how these outcomes fit together within a consistent network architecture.
The RED-SEA network architecture
Architecture, co-design and performance
Optimizing the fit with the other EuroHPC projects and with the EPI processors
- Extending LinkTest for BXI interconnects (blog article)
- RED-SEA: Network Solution for Exascale Architectures (research paper)
- A peek at BXI V2 performance (blog article)
- D1.4 Report on holistic evaluation of RED-SEA network technologies (deliverable)
High-performance Ethernet
Developing a high-performance, low-latency, seamless bridge with Ethernet
- D2.3 RoCE and IPoverBXI Evaluation Report (deliverable)
- Optimized Page Fault Handling During RDMA (research paper)
- D2.9 BXI to Ethernet bridging demonstrator (deliverable)
Efficient Network Resource Management
Including congestion management and Quality-of-Service targets while sharing the platform across application and users
- The APEnet interconnect architecture in RED-SEA (blog article)
- COSSIM framework (blog article)
- The VEF Traces framework (blog article)
Endpoint Functions and Reliability
End-to-end enhancements to network services – from programming models to reliability & security and to in-network compute
- HEAR: Homomorphically Encrypted Allreduce (research paper)
- User-guided Page Merging for Memory Deduplication in Serverless Systems (research paper)
- Cppless: Productive and Performant Serverless Programming in C++ (research paper)
- D4.7 Optimized MPI and compute in network implementations (deliverable)
- Canary: Congestion-aware in-network allreduce using dynamic trees (journal article)
- FaaSKeeper: Learning from Building Serverless Services with ZooKeeper as an Example (research paper)
- Low-latency Communication in RISC-V Clusters (blog article)
- A RISC-V in-network accelerator for flexible high-performance low-power packet processing (research paper – video)
- Flare: flexible in-network allreduce (research paper – video)
- Application-defined, high-performance packet processing with sPIN (blog article)
- A RDMA Interface for Ultra-Fast Ultrasound Data-Streaming over an Optical Link (research paper)
- KafkaDirect: Zero-copy Data Access for Apache Kafka over RDMA Networks (research paper – video)
- Asynchronous Distributed-Memory Triangle Counting and LCC with RMA Caching (conference paper)
- Lifting C semantics for dataflow optimization (research paper)
- D4.3 Planned MPI-related optimizations (deliverable)
- Development of multirail feature in MPC for BXI interconnect (blog article)
- NeVerMore: Exploiting RDMA Mistakes in NVMe-oF Storage Applications (research paper)
- Building blocks for network-accelerated distributed file systems (research paper)
- Influence of Network Performance Variability on Application Scalability (research paper)
- Extending ParaStation MPI by BXI Support (blog article)
- rFaaS: Enabling High Performance Serverless with RDMA and Leases (research paper) – the software prototype, data, analysis scripts, and replication scripts for this paper are available here
- FMI: Fast and Cheap Message Passing for Serverless Functions (research paper)
