Proxied Streaming (PRS)

• Proxied Streaming (PRS) is a paradigm that employs intermediary proxy nodes to bridge data producers and consumers, enhancing performance and security.
• It leverages caching, dynamic load sharing, and cooperative streaming to reduce latency, lower server load, and mitigate network issues like firewall/NAT constraints.
• PRS finds applications in VoD, HPC, live streaming, and ISAC systems, delivering improved throughput and robust, secure streaming workflows.

Proxied Streaming (PRS) is a broad paradigm in which intermediary proxy nodes are deployed between data producers and data consumers to facilitate, optimize, or secure streaming workflows. Though originally used in the context of distributed multimedia systems, the PRS concept has been generalized and adapted across various domains, including high-performance computing (HPC), video-on-demand (VoD), live video streaming, and emerging integrated sensing and communication (ISAC) systems. PRS architectures are characterized by their ability to abstract away network and deployment barriers—such as NAT, firewall restrictions, and WAN inefficiencies—through proxy-mediated tunnels or overlays, often enabling additional functionalities such as cooperative caching, bandwidth aggregation, dynamic load sharing, or application-layer defense mechanisms.

1. Architectural Principles of Proxied Streaming

At its core, Proxied Streaming inserts at least one proxy (or a cascade of proxies) between the source and the sink in a streaming pipeline. In classical VoD systems, proxies reside within “Local Proxy Server Groups” (LPSGs), receiving batched content from central repositories and serving user requests locally or near-locally (0910.1471, Dakshayini et al., 2010, Nair et al., 2010). In cross-facility or HPC scenarios, proxies are deployed at strategic ingress/egress nodes (e.g., Data Transfer Nodes or gateways) to relay high-throughput data streams between edge instruments, intermediary computing tiers, and computational backend facilities (George et al., 28 Sep 2025). In adaptive or cooperative streaming contexts, proxies may further serve as aggregation points, facilitating bandwidth multiplexing or intelligent data delivery paths (Wu et al., 2016).

A general model of PRS includes the following canonical roles:

Role	Example	Function in PRS
Source/Producer	Video server, Edge instrument	Generates/stores stream data
Local Proxy	Edge DTN, VoD PS	Receives/forwards data, handles local access
Remote Proxy	HPC DSN, VoD Tracker	Relays to consumer/facility, additional caching/logics
Consumer/Sink	Client, HPC worker	Endpoint for data consumption/processing

The proxies typically communicate over secure, often mutually authenticated tunnels established via protocols such as TLS (using entities like HAProxy, Nginx, or purpose-built components). Applications communicate only locally with their respective proxies, which abstract away the complexity of cross-domain communication.

2. Caching, Load Sharing, and Cooperative Techniques

A distinctive feature of PRS in multimedia streaming is the use of hierarchical, popularity-driven caching and cooperative streaming. Content is segmented into prefixes and suffixes, with the earliest segments (prefix-1) cached on the proxies to serve the majority of requests rapidly. Upstream trackers or second-tier proxies may cache larger, less frequently accessed segments (prefix-2), reducing dependence on the often distant main servers (0910.1471, Nair et al., 2010). Key allocations follow formulas such as:

• Prefix-1 size for video $i$: $W(\mathrm{pref}\text{-}1)_i = x_i \cdot S_i$
• Prefix-2 size: $$W(\mathrm{pref}\text{-}2)_i = x_i \cdot (S_i - W(\mathrm{pref}\text{-}1)_i)$$

with $x_i$ as the popularity factor, $S_i$ as the total size.

Load sharing algorithms introduce dynamic request routing. If a proxy lacks the requested content locally, it queries the tracker, which can reroute requests to neighbors, adjacent clusters, or—last resort—the main server. Optimal routes are computed to minimize waiting time and transmission cost using constraints such as $M \cdot B = \sum_{i=1}^N W_i$ where $M$ is the number of proxies, $B$ is buffer size per proxy, and $W_i$ is the cached portion (Dakshayini et al., 2010).

In streaming with chaining (editor's term), proxies maintain a list of actively streaming clients. New arrivals within a time threshold relative to the chain leader can join the chain and receive forwardings directly from other clients, amortizing server and network load (Dakshayini et al., 2010, 0910.1471).

3. Performance Evaluation and Comparative Metrics

PRS architectures have been quantitatively evaluated along dimensions including:

• Video Hit Ratio (VHR): Up to 93% of requests served locally with PC+Chaining, compared to 65% without (0910.1471).
• WAN Bandwidth Usage: Substantial reduction, as large fractions of popular content are hit locally or via trackers, minimizing WAN traffic between proxies and central servers (0910.1471). For scientific workflows, PRS provides streaming throughput and RTT within a small factor of direct streaming, even as concurrent consumers scale (George et al., 28 Sep 2025).
• Startup Latency / Client Waiting Time: Caching at proxies yields average initial waiting times of 2–3 seconds, and in synthetic HPC workloads, PRS median RTTs remain under 0.5 s, similar to direct minimal-hop architectures (0910.1471, George et al., 28 Sep 2025).
• Client Rejection Ratio: Lowered significantly, as PRS enables requests to be fulfilled cooperatively and by leveraging local resources (0910.1471, Dakshayini et al., 2010).
• Server Load: Decreases by as much as 70% on the main server thanks to proxy-level offloading (0910.1471, Dakshayini et al., 2010).

In HPC data streaming architectures, PRS demonstrates throughput approaching DTS (Direct Streaming) in broadcast and feedback scenarios, while maintaining superior deployment feasibility compared to Managed Service Streaming (MSS) (George et al., 28 Sep 2025).

4. Proxy-Mediated Streaming for Network and Security Management

Proxies in PRS facilitate not only connectivity but also resilience and system security:

• Firewall/NAT Traversal: Proxies bridge internal and external networks, often through HTTPS/TLS, circumventing address translation and dynamic assignment constraints (George et al., 28 Sep 2025).
• Secure, Encrypted Relays: PRS sessions are generally encapsulated in mutually authenticated TLS tunnels (via Stunnel, HAProxy, etc.), supporting both transport- and app-layer security (George et al., 28 Sep 2025).
• Multipath and Aggregation: Crowdsourced live streaming leverages B-boxes (edge proxy boxes) and MPTCP-enabled cloud proxy servers to aggregate uplinks over disjoint networks, maximizing bandwidth and reliability (Wu et al., 2016).
• DDoS Mitigation: In scalable cloud-based PRS deployments, randomized shuffling and reassignment of clients to proxies (“shuffling defenses”)—sometimes augmented by reputation systems—mitigate attack concentration and restore nominal user service probability, as described by binomial allocation models (Shan et al., 2017).

These capabilities distinguish PRS from flat or direct streaming topologies where administrative configuration and security model may not scale across domains.

5. Application Contexts and Use Cases

PRS is applied across multiple domains, each leveraging proxy roles for differentiated benefits:

• VoD Systems: Caching, prefix segmentation, and chaining for scalable video delivery with minimized request latency (0910.1471, Dakshayini et al., 2010, Dakshayini et al., 2010).
• Scientific Workflows and HPC: Data ingestion and memory-to-memory streaming for real-time analysis and steering of experimental pipelines, including AI-in-the-loop experiments (e.g., GRETA/Deleria and LCLS workflows) (George et al., 28 Sep 2025).
• Crowdsourced Live Streaming: Aggregation boxes and proxy clouds to overcome edge network limitations for uplink-intensive applications (Wu et al., 2016).
• Security-Critical Services: Distributed shuffling to mitigate DDoS attacks and improve benign client availability in multi-tenant, cloud-based streaming systems (Shan et al., 2017).
• Integrated Sensing and Communication (ISAC): Proxy-like mediation of reference signals (e.g., PRS in 5G) enables joint communication, positioning, and sensing in next-generation wireless systems (Wei et al., 2022, Ozbay et al., 18 Jul 2024, Khosroshahi et al., 1 Aug 2024, Khosroshahi et al., 30 Sep 2024).

6. Scalability, Deployment Feasibility, and Research Directions

PRS architectures are inherently scalable due to their decoupling of producer-consumer identity and network addressability:

• Scalability: Proxies deployed at facility ingress/egress nodes can be scaled horizontally. In distributed VoD, ring or cluster topologies (among both proxies and higher-tier trackers) provide scale-out performance and redundancy (0910.1471, Dakshayini et al., 2010, George et al., 28 Sep 2025).
• Feasibility: In production HPC environments, PRS allows rapid deployment over existing infrastructure with only moderate administrative intervention—critical for multi-institution collaborations (George et al., 28 Sep 2025). For very high concurrency, PRS incurs slightly increased overhead but sustains near-constant throughput relative to direct streaming.
• Limitations and Future Work: The main sources of overhead are additional proxy hops, connection handling (particularly with connection managers like Stunnel versus multi-connection proxies like HAProxy), and the need for port and firewall adaptations for public facility deployments. Further research may address adaptive proxy resource scaling, real-time optimization of multi-proxy routing, and integration with autoscaling and stateful session migration for enhanced resilience (Shan et al., 2017, George et al., 28 Sep 2025).

A plausible implication is that PRS, by virtue of its intermediary architecture and flexibility across domains and applications, provides a robust substrate for emerging requirements such as on-the-fly infrastructural security adaptation, real-time data mobility, and hybrid computation/communication in scientific and media networks.

7. Summary Table: PRS Across Workloads

Workload / Domain	Proxy Function	Primary Outcomes
VoD (media streaming)	Caching, client-chaining	Lower latency; reduced server load & bandwidth
HPC/Scientific Data (DS2HPC)	Secure relay, NAT traversal	Near real-time streaming; maintain throughput
Live Streaming (BASS/B-box)	Bandwidth aggregation for uplinks	High-bitrate streaming from edge networks
Cloud Security (Shuffling)	Client-proxy assignment and re-shuffle	DDoS resilience; service continuity
ISAC (5G/6G)	Proxying of reference signals/data	Multipurpose comms and sensing on shared infra

This synthesis captures the architectural scope, operational techniques, quantitative impacts, and development trajectories of Proxied Streaming as derived from detailed experiments and deployments (0910.1471, Dakshayini et al., 2010, Nair et al., 2010, Dakshayini et al., 2010, Wu et al., 2016, Shan et al., 2017, Wei et al., 2022, Ozbay et al., 18 Jul 2024, Khosroshahi et al., 1 Aug 2024, Khosroshahi et al., 30 Sep 2024, George et al., 28 Sep 2025).