ABR ladder design — the engineering decisions in adaptive bitrate streaming

ABR ladder design is the engineering decision that determines what video quality your viewers actually experience. The ladder is the set of (resolution, bitrate, codec) tuples — the rungs — that your encoding pipeline produces and your delivery infrastructure serves. The player picks a rung based on bandwidth, buffer state, and device capability. The ladder you ship is the contract you've made with your viewers about what their experience looks like across the range of their networks and devices. This page is the engineering reference for designing one.

#What an ABR ladder is

An ABR (Adaptive Bitrate) ladder is the set of variant streams produced for a single source. Each rung is a complete encoded stream at a specific (resolution, bitrate, codec). The HLS master playlist or DASH MPD declares the available rungs; the player picks among them.

A typical 2026 premium streaming ladder:

Each rung serves a different combination of bandwidth + device capability. The ladder needs to cover the range of conditions your audience experiences.

#The decisions in ladder design

Several engineering decisions go into a ladder:

Rung count — how many rungs to ship. More rungs = better adaptation granularity, more storage cost, more encoding compute. Typical: 5-8 rungs for premium streaming, 3-5 for budget streaming.

Resolution choices — which specific resolutions to include. 4K, 1440p, 1080p, 720p, 540p, 360p, 240p are the common options. Choose based on audience devices and content nature.

Codec choice per rung — which codec for each rung. Higher tiers favor AV1/HEVC for bandwidth efficiency; lower tiers favor H.264 for universal compatibility.

Bitrate calibration — what bitrate to target at each (resolution, codec) pair. Too low = quality issues; too high = bandwidth waste.

Audio configuration — typically separate audio rungs (stereo AAC at 192 kbps, plus optional 5.1 surround). Audio is multiplexed with video at delivery time but encoded separately.

Subtitle/caption tracks — additional alternate tracks for accessibility. Typically don't impact ABR decision but contribute to manifest complexity.

#Resolution selection

The resolution rungs to include depend on:

Audience device mix — what fraction of viewers have 4K-capable devices? 1080p? Mobile-only? Audience analytics tell you which resolutions matter.

Content nature — premium feature films justify 4K; news content rarely does; UGC content rarely does. Match resolution ceiling to content production value.

Bandwidth constraints in target markets — if your audience is in markets with limited high-bandwidth penetration, the high-resolution rungs are wasted bandwidth in encoding. Conversely, in markets with strong fiber penetration, 4K matters more.

A pragmatic 2026 resolution set for global consumer streaming:

• 4K — top tier for premium content, devices with 4K capability, and bandwidth >= 15 Mbps.
• 1440p — middle-of-pyramid for premium devices in 8-15 Mbps bandwidth range. Sometimes skipped if 4K and 1080p both ship.
• 1080p — universal HD tier. Most audiences end up here.
• 720p — mid-tier fallback. Broad device support, modest bandwidth.
• 540p — mobile/bandwidth-constrained tier.
• 360p — emergency floor for very limited bandwidth or downlevel devices.

Skipping 1440p is common — the rung between 4K and 1080p is often not worth the additional encoding/storage cost. The audience that needs in-between bandwidth typically lands at 1080p or 720p anyway.

#Codec selection per rung

The codec choice per rung depends on audience reach and bandwidth efficiency tradeoffs:

Top tiers (4K, 1440p) — AV1 ideal for AV1-capable audience, HEVC fallback for Apple. AV1 saves ~30% bandwidth vs HEVC for top-tier content.

Middle tiers (1080p, 720p) — HEVC is the workhorse. AV1 for 1080p where audience supports it (saves bandwidth meaningfully). H.264 also at 1080p for downlevel devices that don't decode HEVC.

Floor tiers (540p, 360p) — H.264 only. Bandwidth savings from HEVC/AV1 at this resolution are modest; broad compatibility wins.

The codec mix increases ladder complexity. A "single-codec ladder" (e.g., all HEVC) simplifies pipeline but limits reach. A "multi-codec ladder" (AV1 top + HEVC mid + H.264 floor) requires producing each rung in multiple codec variants.

For 2026 premium streaming, the right answer is multi-codec ladder despite the encoding complexity. The audience reach and bandwidth efficiency benefits compound.

#Bitrate calibration

Bitrate per rung is the most-debated decision. Two approaches:

Universal bitrate ladder — fixed bitrates per (resolution, codec), applied to all content. Simpler operationally; may waste bandwidth on easy content or under-budget hard content.

Per-title bitrate — content-aware bitrate selection (see per-title encoding). Better quality-per-byte but more operationally complex.

For universal ladder design, a starting point for HEVC bitrates (modify based on content type):

• 4K HEVC: 10-15 Mbps
• 1440p HEVC: 5-8 Mbps
• 1080p HEVC: 3-5 Mbps
• 720p HEVC: 1.5-3 Mbps
• 540p HEVC: 800-1500 kbps
• 360p HEVC (rarely used): 400-800 kbps

For H.264 (~50% more bitrate at same quality):

• 1080p H.264: 4.5-7.5 Mbps
• 720p H.264: 2.25-4.5 Mbps
• 540p H.264: 1.2-2.25 Mbps
• 360p H.264: 600-1200 kbps

For AV1 (~30% less bitrate than HEVC at same quality):

• 4K AV1: 7-10.5 Mbps
• 1440p AV1: 3.5-5.6 Mbps
• 1080p AV1: 2.1-3.5 Mbps

These are starting points. Calibrate against your actual content using VMAF — the real bitrate depends on content type, codec preset, and quality target.

#The bandwidth-quality curve

For each (resolution, codec) combo, there's a bandwidth-quality curve. Below a threshold, quality drops sharply. Above a threshold, quality improvements are marginal. The "knee" of the curve is the bitrate sweet spot.

Per-title encoding identifies the knee per content; universal ladders pick a bitrate that's typically slightly above the knee for representative content (so most content has acceptable quality).

VMAF measurement helps calibrate. Pick target VMAF values per tier (e.g., VMAF 93 for top, VMAF 88 for mid, VMAF 80 for floor); for representative content, find the bitrate that achieves each VMAF target; use those as starting points for the ladder.

#Encoder presets

Within the codec choice, the encoder preset matters significantly:

• x264 medium / x265 medium / SVT-AV1 preset 6 — VOD baseline. Slow enough to achieve good quality; fast enough for production throughput.
• x264 fast / x265 fast / SVT-AV1 preset 8 — live encoding. Quality drops modestly; encoding speed increases substantially.
• x264 slow / x265 slow / SVT-AV1 preset 4 — premium VOD. Better quality at the same bitrate; significantly more compute.

Mixing presets across the ladder is occasionally useful — e.g., spend more compute on the top tier (slow preset) and less on the floor (medium preset). Operationally simpler to use one preset across the ladder.

#Audience-mix-driven ladder design

The right ladder for your service depends on your audience:

Premium SVOD (Netflix-like) — large audience, high engagement, premium content. Rich ladder: AV1 top + HEVC middle + H.264 floor. Per-title encoding for bandwidth efficiency.

News and short-form video — broad audience, mobile-heavy, ephemeral content. Simpler ladder: HEVC primary + H.264 floor. Universal bitrates for operational simplicity.

Live sports — large audience, latency-sensitive, action content. HEVC ladder with hardware encode. Lower per-tier bitrates than VOD because hardware encode quality is slightly lower.

UGC platforms — massive scale, varying content quality, varying audience. Aggressive multi-codec strategy with per-content-type rules.

Internal/corporate video — controlled audience, controlled devices. Single-codec ladder (HEVC or AV1) with limited rungs.

Match the ladder to the audience and content. Don't ship a Netflix-style 8-rung multi-codec ladder for an internal corporate communications tool.

#Common ladder design mistakes

The mistakes that are easy to make:

Too many rungs at the top — diminishing returns past 5-6 rungs for most use cases. Extra rungs add operational cost without serving meaningful audience needs.

Skipping the floor — pipelines that ship 720p as their lowest rung exclude bandwidth-constrained viewers. A 360p H.264 floor is cheap and serves the long tail.

Rung gaps that don't allow smooth adaptation — if rung 4 is at 5 Mbps and rung 5 is at 1.5 Mbps, the player jumps too aggressively when bandwidth dips. Closer-spaced rungs adapt smoother.

Per-tier bitrate calibration based on assumptions, not VMAF — bitrates that "feel right" but don't actually hit quality targets. Validate with VMAF.

Ignoring audio bitrates — audio is often left at default (128 kbps stereo AAC) when premium streaming should be at 192 kbps. Audio quality matters more for user experience than most teams realize.

Too many codec variants — every codec variant doubles encoding compute and storage. Triple-codec ladder is the limit for most use cases; quad-codec is rarely justified.

#What MpegFlow does with ABR ladder design

MpegFlow's DAG runtime expresses an ABR ladder as a multi-stage workflow. The partitioner splits the per-rendition encode paths into parallel stages persisted to the job_stages table with dependency tracking. Each rendition runs as its own stage on an FfmpegExecutor worker; a downstream packaging stage depends on all renditions completing. Per-stage retry handles transient failures so a single rendition's OOM doesn't restart the whole ladder. Sibling cancellation on permanent failure stops dependent siblings immediately to avoid wasting compute. Partial-success state preserves completed renditions when one fails non-retryably, with the failed rung surfaced for ops review.

Customers configure the ladder per workflow; default templates per content category (premium VOD, sports, news, internal mezzanine) are starting points to calibrate against your specific content via VMAF measurement (see /topics/quality/abr-ladder-vmaf-calibration).

For customers building the ladder for the first time, the typical process: start with a template appropriate to content category; run it on representative content; compute VMAF on each rung against source; identify rungs that miss quality targets or are over-budgeted; iterate. The template is the starting point, not the destination.

For customers running per-title encoding, the ladder design becomes the candidate pool from which per-title selects content-aware rungs (see /topics/quality/per-title-encoding for the probe → analyze → encode chain that runs as discrete stages with cross-stage data flow).

The workflow YAML lets customers configure the ladder per workflow, so different content types within the same MpegFlow deployment can use different ladders. Different audiences (mobile-first vs premium streaming vs archive) call for different ladders; the same pipeline supports all of them by configuration, not by deployment differences.

ABR ladder design is one of the conversations we have most often during customer onboarding. The ladder is the most consequential single decision in pipeline design — it determines bandwidth costs, perceived quality, and the audience reach. Getting it right pays dividends; getting it wrong is a slow, painful, customer-impacting mistake to fix in production.