Name
Abstract | ||
|---|---|---|
RDMA is becoming prevalent because of its low latency, high throughput and low CPU overhead. However, in current datacenters, RDMA remains a single path transport which is prone to failures and falls short to utilize the rich parallel network paths. Unlike previous multi-path approaches, which mainly focus on TCP, this paper presents a multi-path transport for RDMA,
<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">i.e.</italic>
MP-RDMA, which efficiently utilizes the rich network paths in datacenters. MP-RDMA employs three novel techniques to address the challenge of limited RDMA NICs on-chip memory size: 1) a
<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">multi-path ACK-clocking</italic>
mechanism to distribute traffic in a congestion-aware manner without incurring per-path states; 2) an
<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">out-of-order aware path selection</italic>
mechanism to control the level of out-of-order delivered packets, thus minimizes the meta data required to them; 3) a
<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">synchronise</italic>
mechanism to ensure in-order memory update whenever needed. With all these techniques, MP-RDMA only adds 66B to each connection state compared to single-path RDMA. Our evaluation with an FPGA-based prototype demonstrates that compared with single-path RDMA, MP-RDMA can significantly improve the robustness under failures (
<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$2\times \sim 4\times $ </tex-math></inline-formula>
higher throughput under 0.5%~10% link loss ratio) and improve the overall network utilization by up to 47%. |
Year | DOI | Venue |
|---|---|---|
2019 | 10.1109/TNET.2019.2948917 | IEEE/ACM Transactions on Networking |
Keywords | Field | DocType |
Datacenter networks,RDMA,multi-path transport,hardware-based transport | Multi path,Computer science,Computer network,Remote direct memory access | Journal |
Volume | Issue | ISSN |
27 | 6 | 1063-6692 |
Citations | PageRank | References |
0 | 0.34 | 0 |
Authors | ||
8 | ||
Name | Order | Citations | PageRank |
|---|---|---|---|
| Guo Chen | 1 | 11 | 3.96 |
| Yuanwei Lu | 2 | 35 | 2.66 |
| Bojie Li | 3 | 63 | 5.85 |
| Kun Tan | 4 | 1350 | 98.64 |
| Yongqiang Xiong | 5 | 708 | 45.84 |
| Peng Cheng | 6 | 31 | 5.56 |
| Jiansong Zhang | 7 | 596 | 48.98 |
| Thomas Moscibroda | 8 | 4047 | 200.40 |