~/talk

State-Compute Replication: Parallelizing High-Speed Stateful Packet Processing

Talkat NSDI '25 · 2025-04-28..04-30· [networking, distributed-systems]
links:none on file
links:none on file
name:Tao Wang
links:none on file
links:none on file
links:none on file
links:none on file
links:none on file
slides:none on file

Introduces state-compute replication that replicates only the state-update code and attaches compact per-packet histories so each core fast-forwards private state without locks, implemented as a packet history sequencer on programmable NIC/switch hardware. Prototype uses XDP applications and hardware described as Tophino and net fpj, stores histories in an on‑chip ring buffer, and shows near-linear throughput scaling with cores across three real traces and even a single large flow. Limits appear when state compute dominates CPU or when bandwidth caps are hit (token bucket stalls ~11 cores).

AI-written summary · about

  1. 0:00State-compute replication intro
  2. 1:52Scaling stateful packet processing
  3. 3:39Sharing and sharding approaches
  4. 5:27SCR overview
  5. 7:13Packet processing time breakdown
  6. 8:17SCR with programmable NICs
  7. 10:03Sequencer hardware design
  8. 11:27SCR performance evaluation
  9. 13:17SCR limitations and bottlenecks
  10. 14:41Loss recovery in SCR
  11. 15:25Conclusion and benchmarks
~/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.