Topics proposed for IEEE 802 1 20 0002

Background • Nendica has considered proposing a new Work Item on ‘Revision of “The

Challenges in today’s Data Center • RDMA expanding to new Applications • Distributed Storage(e.

Potential topics Contribution driven, so content depends on inputs New topics potentially include: •

Potential challenges to be considered • Data Center New Challenges in the future •

References 1. https: //datatracker. ietf. org/doc/draft-zhuang-tsvwg-ai-ecn-for-dcn/ 2. https: //mentor. ieee. org/802. 1/dcn/20/1 -20 -0002

Slides: 6

Download presentation

Topics proposed for IEEE 802. 1 -20 -0002 -01 Marcus Sun Huawei 2020 -02 -05

Background • Nendica has considered proposing a new Work Item on ‘Revision of “The Lossless Network for Data Centers”’ • IEEE 802. 1 -20 -0002 -01 https: //mentor. ieee. org/802. 1/dcn/20/1 -20 -0002 -01 -ICne. pdf • Slide 8 of the draft proposal (“Potential topics ”) has open items • Contribution driven, so content depends on inputs • New topics potentially include: • (tbd prior to submitting to 802. 1/802) • This contribution proposes to complete the tbd on Slide 8, per Slide 4 • Slide 3 and 5 identifies some challenges for today and future’s Data Centers

Challenges in today’s Data Center • RDMA expanding to new Applications • Distributed Storage(e. g. NVMe over Fabric) • Distributed Computing(e. g. HPC/AI Training) • Network Challenges for RDMA applications • • • High I/O throughput with low-latency storage network Ultra-low latency network for distributed computing Bandwidth vs. Latency tradeoff Congestion Control is critical in large-scale network Congestion Control parameters challenging to optimize (e. g. Priority Flow Control (PFC) threshold)

Potential topics Contribution driven, so content depends on inputs New topics potentially include: • Approaches to PFC storms elimination • • • Improving Congestion Notification • • Deadlock detection Deadlock elimination Issues with Explicit Congestion Notification Enhanced version of Quantized Congestion Notification (originally IEEE 802. 1 Qau) Feasibility of including Qo. S support congestion parameter optimization • • heuristic algorithm for identifying congestion parameters Methods for dynamic optimization based on services 4

Potential challenges to be considered • Data Center New Challenges in the future • Applications Challenges • Compute Centric • Memory Fabric • Any New Opportunities for Layer 1/2?

References 1. https: //datatracker. ietf. org/doc/draft-zhuang-tsvwg-ai-ecn-for-dcn/ 2. https: //mentor. ieee. org/802. 1/dcn/20/1 -20 -0002 -01 -ICne. pdf