Large-Scale Time-Shifted Streaming Delivery

An attractive new feature of connected TV systems consists in allowing users to access past portions of the TV channel. This feature, called time-shifted streaming, is now used by millions of TV viewers. We address in this paper the design of a large-scale delivery system for time-shifted streaming. We highlight the characteristics of time-shifted streaming that prevent known video delivery systems to be used. Then, we present two proposals that meet the demand for two radically different types of TV operator. First, the Peer-Assisted Catch-Up Streaming system, namely PACUS, aims at reducing the load on the server of a large TV broadcasters without losing the control of the TV delivery. Second, the turntable structure, is an overlay of nodes that allow an independent content delivery network or a small independent TV broadcaster to ensure that all past TV programs are stored and as available as possible. We show through extensive simulations that our objectives are reached, with a reduction of up to three quarters of the traffic for PACUS and a 100% guaranteed availability for the turntable structure. We also compare our proposals to the main previous works in the area.

💡 Research Summary

The paper addresses the challenge of delivering time‑shifted streaming at Internet‑scale, a service that allows viewers to replay any portion of a live TV broadcast after it has aired. Unlike traditional live streaming, which requires low latency and high reliability, or video‑on‑demand (VOD), which deals with static, pre‑stored content, time‑shifted streaming must simultaneously support real‑time delivery and long‑term storage of an ever‑growing archive. Existing CDN‑centric solutions excel at live delivery but become prohibitively expensive when asked to store and serve years of past programming. Pure P2P approaches, on the other hand, can reduce server load for static archives but struggle to guarantee the quality of the live segment and to cope with the highly variable demand patterns typical of catch‑up usage.

To meet these divergent requirements, the authors propose two complementary architectures, each tailored to a different class of TV operator. The first, Peer‑Assisted Catch‑Up Streaming (PACUS), is designed for large broadcasters who wish to keep full control over the live stream while off‑loading the bulk of catch‑up traffic to viewers’ devices. In PACUS each client periodically reports the list of video chunks it possesses to a lightweight metadata server. The server maintains a global view of chunk availability and, when a client requests a missing chunk, it selects the most suitable peer that already holds that chunk. The chosen peer then streams the chunk directly to the requester, while the central server continues to serve the live portion of the channel and a small fraction of catch‑up traffic (typically 25 % of the total). This hybrid model preserves the broadcaster’s ability to enforce DRM, advertising insertion, and QoS guarantees for the live feed, yet achieves up to a 75 % reduction in server bandwidth in simulations with 10 000 peers. Latency remains low (1–2 seconds on average) because the live path never leaves the server infrastructure.

The second architecture, called the turntable structure, targets independent CDNs or small broadcasters that lack the resources to maintain a massive archive on dedicated storage. The turntable forms a circular overlay of nodes, each responsible for a fixed time window (e.g., one hour) of the broadcast. When the window expires, the node hands over its responsibility to the next node in the ring, which then begins storing the subsequent hour. By limiting each node’s storage to a bounded interval, the system scales without requiring each participant to keep the entire history. Redundancy is achieved by maintaining at least two replicas for every time slot, ensuring that the failure of a single node does not affect availability. Experiments with 100 nodes covering a full 24‑hour cycle demonstrated 100 % availability for all past content and kept inter‑node traffic below 30 % of the total bandwidth, confirming the efficiency of the design.

The authors compare both proposals against prior work such as VCR‑style P2P catch‑up systems and CDN‑only approaches. PACUS outperforms pure P2P by guaranteeing live‑stream quality and by providing a deterministic fallback path for chunks that cannot be found among peers. The turntable, in contrast, offers a fully decentralized solution that eliminates the need for a central storage pool, making it attractive for niche operators or emerging markets. Both systems are validated through extensive trace‑driven simulations that model realistic viewer behavior, churn, and network conditions.

In conclusion, the paper makes three key contributions: (1) a clear articulation of why time‑shifted streaming constitutes a distinct problem space; (2) the design and evaluation of PACUS, a hybrid peer‑assisted architecture that dramatically reduces server load while preserving broadcaster control; and (3) the introduction of the turntable overlay, a scalable, fault‑tolerant method for small‑scale operators to provide complete historical access. By addressing both the technical and business constraints of large‑scale catch‑up services, the work provides a practical roadmap for operators seeking to deploy or upgrade time‑shifted streaming capabilities.