Skip to content

PATTERN Cited by 1 source

Nominal-bitrate admission control

Problem

When a server admits variable-bitrate (VBR) sessions based on current observed traffic, a long low-bitrate phase (easy-scene content) tricks the admission controller into seeing spare headroom and admitting more sessions than the link can actually carry at worst case. When the content turns complex and all admitted sessions spike simultaneously toward their nominal rates, the aggregate traffic exceeds what the network link or NIC can sustain; latency rises, packets drop, and devices stall or quality- downshift (Source: sources/2026-04-02-netflix-smarter-live-streaming-vbr-at-scale):

"During a slow, easy sequence, the encoder might only generate 2 Mbps for the 5 Mbps stream — or even less — to maintain its target quality, and it can stay at that lower level for an extended period. The server then appears to have plenty of unused capacity: per-session bitrate is low and aggregate traffic is well below its limits. Our steering systems naturally interpret this as a signal that the server is under-utilized and can accept more sessions… If the server has admitted many additional sessions during the preceding low-bitrate period, the aggregate traffic can suddenly exceed what the network link or NIC can sustain."

This is the core reason current-traffic-as-capacity-proxy admission control is unsafe for VBR.

Solution

Reserve capacity against the stream's nominal bitrate (equivalently: the capped-VBR cap for each stream), not its current observed bitrate. Treat every admitted VBR session as if it could snap back to its nominal rate at any moment. Under capped VBR, this is safe by construction because the cap is a hard instantaneous ceiling (Source: sources/2026-04-02-netflix-smarter-live-streaming-vbr-at-scale):

"Our fix was to change how we decide whether a server can take more sessions. Instead of basing that decision only on current traffic, we reserve capacity based on each stream's nominal bitrate, not just what it happens to be using at that moment. Even if a VBR stream is currently in a very cheap, low-bitrate phase, we still treat it as something that can quickly return to its nominal rate."

What this preserves

  • CBR-equivalent admission-control safety. The admission controller now makes the same decisions it did under CBR — reserve N × nominal per admitted session — so fleet-steering behaviour doesn't change across the rate-control cutover.
  • VBR efficiency gain on bytes. The admitted sessions still emit fewer bytes on average (≈15% at Netflix), so peak-minute traffic also drops (≈10% at Netflix) even though each session is counted at nominal for admission.
  • Operational consistency across the fleet. Traffic- steering logic doesn't need two modes for two rate-control settings.

What this trades away

  • Conservative admission. Nominal is the worst case; almost no moment has every stream simultaneously at nominal. So the server has "additional headroom" that current-traffic-based admission would have filled. Under nominal-reservation, that headroom is unused on average.
  • No statistical-multiplexing benefit on admission. N independent streams rarely all peak together; the sum-of-nominals massively over- estimates the 99th-percentile aggregate.

Netflix's forward-work direction is to recover that headroom via a measurement-informed "discount" on the nominal-reservation (Source: sources/2026-04-02-netflix-smarter-live-streaming-vbr-at-scale):

"We are experimenting with making our capacity reservation less conservative. Today we reserve based on nominal bitrates to keep servers safe; by carefully applying a 'discount' informed by real VBR behavior, we hope to free up additional headroom without sacrificing stability."

The natural shape is a probabilistic aggregation (what's the 99th-percentile of sum(VBR_session_i)?) combined with a safety margin — classic statistical-multiplexing sizing on top of the conservative baseline.

When to apply

  • Any admission controller that admits variable-bitrate flows (video, audio, mixed-media) where the flow's instantaneous rate can be materially below its worst-case rate for extended periods.
  • CDN fleet steering for streaming video after a CBR → VBR cutover.
  • Admission control for any service with multi-modal load (e.g. queries that run cheap most of the time but occasionally run expensive) where current-load as admission proxy is unsafe under correlated spikes.

Generalisation beyond video

The structural framing generalises: when load is multi-modal or temporally correlated across admitted sessions, current-load-based admission control is unsafe and should be replaced with worst-case-reservation-based admission control. The cost is lower average utilisation; the benefit is that fleet-safety invariants don't depend on uncorrelated-load assumptions that can fail.

Seen in

Last updated · 319 distilled / 1,201 read