A high-performance container overlay network enabling RDMA communication and accelerating TCP to bare-metal performance without application modifications.
Freeflow is a high-performance container overlay network that enables RDMA communication and accelerates TCP sockets to host network performance levels. It solves the problem of network performance degradation in containerized environments by providing bare-metal-like speeds without modifying application code.
System administrators, DevOps engineers, and researchers deploying high-performance computing (HPC) or data-intensive applications (like Spark, TensorFlow) in containerized, multi-tenant cloud or cluster environments.
Developers choose Freeflow because it delivers RDMA and accelerated TCP performance in containers with zero application code changes, integrates with existing overlay networks, and offers flexible operation modes tailored to isolation and efficiency needs.
High performance container overlay networks on Linux. Enabling RDMA (on both InfiniBand and RoCE) and accelerating TCP to bare metal performance. Freeflow requires zero modification on application code/binary.
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Enables RDMA over InfiniBand and RoCE in containers, delivering high throughput and low latency as shown in performance tests with Spark and TensorFlow.
Accelerates TCP sockets to host-level performance, achieving up to 25Gbps throughput and sub-20 microsecond latency per the README's TCP mode description.
Works transparently by intercepting APIs, allowing existing applications like Nginx or Kafka to run without modification, as stated in the applications section.
Offers fully-isolated, semi-isolated RDMA, and TCP modes to balance isolation and efficiency for different cluster types, such as multi-tenant clouds or single-tenant setups.
Requires downloading specific RDMA drivers, editing hard-coded IP addresses in source code, and executing multiple Docker commands, making deployment non-trivial and error-prone.
Fully-isolated RDMA mode incurs a CPU penalty for better isolation, which could impact performance for CPU-sensitive applications despite high network throughput.
Explicitly labeled as a research prototype with no production quality guarantees, limiting reliability and support for critical environments.