~/talk

Approximating Fair Queueing on Reconfigurable Switches

Talkat NSDI '18 · 2018-04-09..04-11· [networking, distributed-systems]
links:none on file
name:Ming Liu
links:none on file
links:none on file
slides:none on file

Implementation of approximate Fair Queueing on reconfigurable switches uses per-packet processing and mutable on-switch state to emulate bit-by-bit round robin with a count-min sketch variant for per-flow rounds, a bytes-per-round quantum, and a Rotating Strict Priority dequeue implemented via a small set of FIFO queues. Prototype on a Cavium NPU and simulations on a leaf-spine topology show using 12–16 FIFOs and <10% extra resources yields 2–4x faster average short-flow completion and 4–8x better 99th-tail latency versus TCP/DCTCP.

AI-written summary · about

  1. 0:00Intro
  2. 0:37Congestion Control Today
  3. 1:32Fair Queueing : in-network enforcement
  4. 2:46Fair Queueing without per-flow queues
  5. 6:53Realizing Fair Queueing on Reconfigurable Switches
  6. 7:59Our approach: Approximate Fair Queueing
  7. 8:41Storing Approximate Flow Counters
  8. 11:28Buffering Packets in Approximate Sorted Order
  9. 13:09Rotating Strict Priority (RSP)
  10. 13:54Realizing an RSP Scheduler
  11. 15:04Summary of Techniques
  12. 17:09Evaluation
  13. 17:21Prototype Implementation
  14. 19:08Simulation Results: Comparison to Fair Queueing
~/search
type to search 6,249 conferences, 138,987 speakers and 185,172 talks
~/help
cmdctrl k
search every conference, speaker and series
/
focus the filter search
h
home
c
conferences
s
speakers
e
series
t
light / dark — the status bar button does the same
?
this help
esc
close this, then clear every filter — or just the search, while you are typing in it

Ignored while you are typing. Pages without a filter rail have no search to focus.