Co-Designing Traffic Control with NVMe-oF for Disaggregated Storage: A Comparative Study
summary
Compares switched and switchless NVMe-oF SANs using small real prototypes and large simulations, NVMe over TCP, BlueField-2 SmartNICs, and Samsung PM983 SSDs; implements INT-assisted symmetric routing, eager bandwidth reservation, and storage-driven traffic scheduling. Throughput matches between architectures, average read latency falls ~28% and tail latency ~25% with switchless in YCSB+RocksDB tests, and switchless shows ~50.5% capital savings at 1,000 ports by avoiding high-radix switches.
AI-written summary · about
chapters (14)
- 0:00Storage disaggregation research
- 0:40Storage disaggregation benefits
- 1:45Storage node trends
- 3:52NVMe over Fabrics traffic
- 8:04Traffic control limits
- 8:25Co-design principles for NVMe-oF
- 9:08INT symmetric routing
- 11:18Eager bandwidth reservation
- 12:24Storage-driven traffic scheduling
- 13:28Switched vs switchless SAN
- 14:54Latency and throughput results
- 15:15Switchless SAN cost savings
- 15:57NVMe-oF observations and solutions
- 16:41Conclusion
related talks
30mFrom ISCSI to NVMe-of: A Comparative Look on Storage Protocol Advantages & Challenges
15mCharacterizing Off-path SmartNIC for Accelerating Distributed Systems
17mRearchitecting the TCP Stack for I/O-Offloaded Content Delivery
26mApplying New High-Throughput Network Solutions in IEC (Integrated Edge Cloud)
47mThe World of 100G Networking
23mFast Software Cache Design for Network Appliances
16mA High-Speed Stateful Packet Processing Approach for Tbps Programmable Switches
01h 24mPreview Session