[FS_QStream_MED] Application Scenarios
This contribution addresses the FS_QStream_MED Rel-20 study item, which evaluates whether current and future media services could benefit from QUIC-based streaming technologies compared to TCP-based streaming (HTTP 1.1 and HTTP/2). The document proposes five application scenarios with their delivery characteristics to be considered in the study.
Description:
Traditional segmented on-demand streaming where UX is driven by TTFF, rebuffering avoidance, and steady-state quality. Serves as a baseline to evaluate whether HTTP/3 improves startup robustness, reduces rebuffering under loss, and stabilizes bitrate selection without changing application semantics or CDN caching.
Delivery Characteristics:
- Latency target: Not live-edge constrained (focus on startup delay and seek response)
- Segment/chunking: Typically 2–6s segments, usually no chunking
- Session duration: Long (tens of minutes to hours)
- Churn: Low (few items per session, seeking within session)
- Cacheability: Very high (content reused across many viewers and time)
- Traffic pattern: Downlink-dominant
Description:
One-to-many distribution with very high fanout where edge replication and caching efficiency are critical. Downlink includes synchronized consumption of common live timeline, significant join/rejoin activity, and sensitivity to end-to-end latency drift. Central for comparing DASH over HTTP/3 versus other QUIC-based options in terms of join time, live latency distribution, stability under congestion/loss, and operational scalability.
Delivery Characteristics:
- Latency target: ~15–45+ seconds (service dependent)
- Segment/chunking: Typically 6–10s segments, usually no chunking (or chunking not required)
- Session duration: Long (tens of minutes to hours)
- Churn: Low (infrequent zaps vs. short-form)
- Cacheability: High (segments cache-friendly, many viewers request same objects)
- Traffic pattern: Downlink-dominant (uplink mostly control/telemetry)
Description:
Live services with tight latency budgets (e.g., interactive sports, live commerce, auctions) where tail latency and timeliness are as important as average latency. Uses very small chunks/parts and constrained playback buffers. Primary service risk is not only rebuffering but also late delivery causing latency drift or missed deadlines. Main stress test for QUIC-based approaches, highlighting where HTTP/3's transport properties may improve QoE and where alternatives like WebTransport or MoQ-style object delivery might reduce overhead and improve timeliness under loss and mobility.
Delivery Characteristics:
- Latency target: ~2–8s (sometimes ~1–2s in aggressive "ultra-low" configurations)
- Segment/chunking: Often ~1–2s segments plus CMAF chunks/partial segments of ~0.4–1.0s
- Session duration: Medium to long (minutes to hours, event-driven)
- Churn: Medium (more joins/leaves than linear, less than short-form)
- Cacheability: Moderate to high (finer chunking increases request/object rate, may reduce cache hit efficiency)
- Traffic pattern: Downlink-dominant (uplink mainly interaction/chat signals)
Description:
Characterized by extremely high churn with short sessions and frequent abandonment within seconds. Platform relies heavily on prefetch and fast startup. Downlink traffic is bursty with many small or partial transfers, frequent representation switches, and strong dependence on connection reuse and efficient request scheduling. Dominant QoE driver is TTFF rather than steady-state quality. Relevant to quantify QUIC's potential benefits in setup/handshake amortization, reduced head-of-line effects, improved performance under lossy access networks, and reduced per-request overhead.
Delivery Characteristics:
- Latency target: Primarily instant start (TTFF is critical "latency" KPI)
- Segment/chunking: Very small initial fetches (tiny first segment/chunk/GOP-aligned unit) followed by progressive fetch; segment durations ~1–2s with emphasis on early playable data
- Session duration: Variable (often minutes, composed of many short items)
- Churn: Very high (many joins/exits per minute)
- Cacheability: Mixed (popular clips cache well, but personalization and huge catalog reduce cache hit ratios)
- Traffic pattern: Downlink-dominant (plus uplink for telemetry, recommendation signals, occasional UGC upload)
Description:
Couples challenging uplink contribution path with large-scale downlink distribution, reflecting modern "creator live" streaming services. Uplink is often mobile and highly variable with fluctuations in available bitrate, RTT, loss, and NAT rebinding. Downlink requires scalable fanout to heterogeneous receivers with frequent joins/rejoins and sensitivity to stalls and latency drift. End-to-end performance frequently dominated by recovery behavior when uplink path changes. Particularly relevant to assess QUIC features supporting robustness and continuity across path changes and to evaluate system-level KPIs such as recovery time after impairment, join/rejoin success rate, and end-to-end latency distributions under mobility.
Delivery Characteristics:
- Latency target: Often sub-second to a few seconds contribution delay (compounds into end-to-end latency)
- Segment/chunking: Can be chunked (frame/GOP/fragment level) or message/object oriented; if segmented ingest, segments tend to be small and/or chunked to reduce contribution delay
- Session duration: Medium to long (minutes to hours)
- Churn: Medium (streams start/stop; less rapid than short-form)
- Cacheability: Low in uplink leg (ingest traffic unique per stream), though downstream distribution may be cacheable
- Traffic pattern: Uplink-dominant on contribution leg (downlink for monitoring/return video is secondary)
The contribution proposes to document all five application scenarios described in section 2 in the new TR for FS_QStream_MED.