Ilmu Komputer & AI editorial
Performance Evaluation of RED-ONION: A High-Speed Disk-to-Disk Transfer System
The core problem
Modern experimental instruments generate data faster than general-purpose file transfer interfaces can move it, making delivery to computing infrastructure a bottleneck in the research process. At many universities and research institutes, the instruments that produce research data and the high-performance computing (HPC) systems that analyze it are separated both geographically and organizationally, because each demands its own expertise and installation environment. Seamlessly connecting the two is a pressing challenge for data-driven science.
This article presents RED-ONION, a high-speed disk-to-disk transfer system that connects research facilities, on campus and beyond, to a computing center. The system combines data transfer nodes (DTNs), a dedicated high-bandwidth network, an all-flash parallel file system, and multi-threaded transfer software that parallelizes network transmission and storage access. The design targets the wire rate both along the entire path—from the read on the sender storage to the write on the receiver storage—and for a single file between one pair of nodes rather than only in aggregate over many files or nodes. The authors describe the end-to-end opti
Innovation
The prototype deployed over a 100 Gbps transpacific path between Atlanta and Tokyo with a 150 ms RTT achieved a single 1 TB file transfer at 90 Gbps. This delivered a terabyte in approximately 95 seconds. The achieved throughput represents 90% of the 100 Gbps wire rate, demonstrating that the end-to-end optimizations across transfer software, operating system, and storage are effective even over a high-latency, high-bandwidth path.
The key quantitative results are summarized below:
| Metric | Value |
|--------|-------|
| Network capacity | 100 Gbps |
| Round-trip time (RTT) | 150 ms |
| File size | 1 TB |
| Achieved throughput | 90 Gbps |
| Transfer time | ~95 s |
| Wire rate utilization | 90% |
These results are for a single file between one pair of nodes, not an aggregate over many files or nodes. This is a stronger result because it demonstrates that a single stream can saturate the link, which is important for applications that need to move large individual files quickly. The transfer time of approximately 95 seconds for 1 TB means that moving a dataset of this size becomes a routine step, and the computing center can serve an instrument as if the two were co-located.
The a
Why it matters
The RED-ONION results demonstrate that disk-to-disk transfer at near-wire speed is achievable over long-distance, high-bandwidth networks when the entire path is optimized. The 90 Gbps achieved over a 100 Gbps link with 150 ms RTT represents 90% wire rate utilization for a single file between one pair of nodes. This is significant because many existing transfer tools achieve high aggregate throughput only when using many parallel streams or many files, but struggle to saturate a link with a single stream over high-latency paths.
The key enabling factors are the combination of data transfer nodes, a dedicated high-bandwidth network, an all-flash parallel file system, and multi-threaded transfer software. The multi-threaded software parallelizes network transmission and storage access, which is essential for hiding the 150 ms RTT and keeping the network pipe full. The all-flash parallel file system provides the storage bandwidth to feed the network, and the dedicated network ensures no competing traffic. The end-to-end optimizations across software, OS, and storage are necessary because a bottleneck at any stage would limit the overall throughput.
The implications for data-driven science are profound. At many universities and research institutes, instruments and HPC systems are separated geographically and organizationally. RED-ONION connects them seamlessly, allowing the computing center to serve an instrument as if the two were co-located. Moving a 1 TB dataset becomes a routine step, which changes the workflow of researchers: they can generate data at the instrument, transfer it to the computing center in about 95 seconds, and analyze it there without worrying about the transfer being a bottleneck.
The limitations of the study include that the evaluation is on a prototype and only for a single file size (1 TB) and a single path (Atlanta–Tokyo, 100 Gbps, 150 ms RTT). Future work could explore other file sizes, multiple concurrent transfers, different network conditions, and integration with existing data transfer tools. The taxonomy candidates for this work include Architecture, Cybersecurity, Network, and Cryptography, though the primary focus is on network and architecture.
In conclusion, RED-ONION shows that with end-to-end optimization, disk-to-disk transfer can achieve 90% of wire rate over a 100 Gbps transpacific path with 150 ms RTT, delivering a terabyte in approximately 95 seconds. This makes large-scale data transfer a routine step and enables the computing center to serve instruments as if they were co-located.
Who should read this
Opening member content…