Everything Matters in Programmable Packet Scheduling
summary
Presents PACKS, a quantile-based admission and queue-mapping design that approximates both PIFO queues scheduling and admission behaviors using a sliding-window rank distribution and per-queue occupancy. Fully implemented in P4 and evaluated in packet-level simulation and on Intel Tofino hardware, PACKS runs at line rate, reduces rank inversions by up to 7× and 15× versus SP-PIFO and AIFO, cuts packet drops up to 60% versus SP-PIFO, and lowers mean FCT for small flows by up to 33% and 2.6×.
AI-written summary · about
chapters (11)
- 0:00Programmable packet scheduling intro
- 1:27Why it matters
- 2:51PyQ abstraction behaviors
- 4:16Limitations of prior PyQ approx
- 7:03Admission and scheduling matter
- 8:26PGX approximates PyQ admission
- 10:11Real-time adaptation in PGX
- 11:36PGX in a single slide
- 12:17Evaluating PGX in software/hardware
- 13:01Key results: drops and inversions
- 14:44Impact on flow completion times
related talks
23mSP-PIFO: Approximating Push-In First-Out Behaviors using Strict-Priority Queues
13mSifter: An Inversion-Free and Large-Capacity Programmable Packet Scheduler
22mEiffel: Efficient and Flexible Software Packet Scheduling
22mProgrammable Calendar Queues for High-speed Packet Scheduling
16mGearbox: A Hierarchical Packet Scheduler for Approximate Weighted Fair Queuing
19mNet-P4ct: Enhanced WAN Bandwidth Fair Sharing Using P4 Programmable Switches
16mPacket Order Matters! Improving Application Performance by Deliberately Delaying Packets
16mManaging Congestion Control Heterogeneity on the Internet with Approximate Performance...
