Stateless Network-Aware Adaptive Bitrate Streaming over IPFS Iliya Mirzaei∗
Shabnam Jafarzade Mojaveri∗
Amirhossein Najafizadeh∗
Stony Brook, NY, USA [email protected]
Stony Brook, NY, USA [email protected]
Stony Brook, NY, USA [email protected]
arXiv:2606.29574v1 [cs.NI] 28 Jun 2026
Abstract Modern content delivery systems are increasingly adopting decentralized architectures to improve availability, reduce operational costs, and better serve geographically distributed users. The InterPlanetary File System (IPFS) represents a promising approach by using content-based identifiers distributed across a global peer-topeer network. Although IPFS improves performance, fault tolerance, resilience, and censorship resistance, its inherently unpredictable environment introduces significant performance variability that limits the effectiveness of conventional Adaptive Bitrate (ABR) streaming mechanisms and degrades Quality of Experience (QoE). Recent network-aware ABR solutions have emerged to address this challenge by incorporating IPFS-specific information into bitrate selection decisions. However, these approaches rely on maintaining continuously synchronized state across consumers and providers during content delivery, which can quickly become stale in the presence of peer churn, provider migrations, network partitions, and changing content distributions. As a result, existing networkaware ABR policies become less effective. In this paper, we investigate whether network-aware ABR approaches can remain effective without maintaining synchronized adaptation state. To this end, we present a stateless network-aware ABR policy for IPFS-based video streaming. Our approach replaces provider-stateful adaptation with an observation-driven policy that recomputes bitrate decisions for each segment using only locally observable request-time signals. To preserve end-to-end adaptation context without maintaining synchronized state at providers, the client embeds its adaptation state in HTTP headers, ensuring that state remains under client control and is carried transparently across requests. By eliminating cross-provider state synchronization, the framework improves robustness to system failures and network reconfigurations while simplifying deployment at scale. Early results show the approach achieves up to roughly 6× higher QoE than a non-adaptive, uncached configuration in our testbed. These findings demonstrate that stateless network-aware adaptation provides a practical and scalable foundation for decentralized video delivery.
Keywords Adaptive bitrate streaming, InterPlanetary File System (IPFS), decentralized content delivery, video streaming, Quality of Experience (QoE), peer-to-peer systems, network-aware adaptation
1
Introduction
Content delivery is shifting from centralized origins toward decentralized architectures that improve availability, lower operational cost, and place content closer to geographically distributed users [19, 24, 39]. Recent years have seen growing interest in decentralized content delivery platforms such as the InterPlanetary ∗ These authors contributed equally to this work.
File System (IPFS) [4, 30]. Such systems are particularly attractive in environments where centralized platforms are unavailable, restricted, or difficult to trust. Studies of alternative software distribution ecosystems show that users frequently rely on decentralized or third-party content sources when access to official distribution channels is limited [23]. IPFS replaces traditional location-based addressing with cryptographic content identifiers (CIDs) and distributes content across a global peer-to-peer network. Content can be replicated and served by multiple independent providers. Therefore, users can retrieve data from any reachable replica rather than a single origin server [30]. This architecture improves availability, fault tolerance, and geographic reach while reducing dependence on centralized infrastructure [30]. Video streaming is among the most demanding content delivery workloads, driving the need for highly scalable and geographically distributed content delivery infrastructures [32]. The decentralized design of IPFS offers an appealing alternative to traditional CDN systems by enabling content replication across a global peer-to-peer network without relying on centralized servers [27]. In principle, these properties can improve availability, reduce infrastructure costs, and increase resilience to failures [32]. In practice, however, they introduce substantial variability into the content delivery process [40]. Since content may be served by different providers and traverse different network paths from one request to the next, the performance seen by a streaming client becomes significantly less predictable than in conventional CDN systems. State-of-the-art content video streaming systems use HTTPadaptive bitrate (ABR) streaming which encodes content at several bitrates and the client continually selects the highest bitrate it believes the network can sustain [14]. ABR algorithms highly depend on a stable and predictable estimate of available throughput [21, 25, 35, 41]. With traditional CDN, consecutive video segments are often served from the same cache or a small set of nearby servers, making past throughput measurements useful predictors of future performance [38]. In IPFS, however, each segment request may be satisfied by a different provider, traverse a different network path, or experience a different lookup delay. Many factors such as host availability, content placement, and content retrieval paths can change rapidly [37], causing large fluctuations in latency and throughput. As a result, a throughput sample observed for one segment may be a poor predictor of the next, leading conventional ABR algorithms to oscillate between bitrates or incur playback stalls. Network-aware ABR systems have emerged to address this challenge [6, 29, 40]. These systems improve Quality of Experience (QoE) by incorporating specific network information into bitrate selection. However, they rely on maintaining state, such as throughput histories and provider performance statistics, across clients, providers, and intermediary proxies. In decentralized environments, maintaining such state is inherently difficult [20]. Peer churn, provider
Iliya Mirzaei, Shabnam Jafarzade Mojaveri, and Amirhossein Najafizadeh
migration, network partitions, and shifting content placement can quickly invalidate previously collected information, causing adaptation decisions to be based on stale observations. Moreover, synchronizing state across distributed components introduces additional complexity and creates scalability bottlenecks as the number of clients, providers, and proxy instances grows. This raises a fundamental question: Can network-aware ABR remain effective in IPFS without relying on maintained state? To answer this question, we present a soft-stateless networkaware ABR policy for IPFS video streaming. Rather than maintaining long-lived synchronized state, our approach makes bitrate decisions independently for each segment using only information available at request time, including recent segment throughput, cache hit or miss status, gateway and node response latency, and client buffer occupancy. Any state retained is strictly local and short-lived, existing only long enough to inform the current adaptation decision. Early results demonstrate that the proposed approach achieves up to roughly a 6× higher QoE than a non-adaptive, uncached configuration in our testbed.
2
Background
Decentralized Content Delivery. IPFS stores data as a Merkle DAG of content-addressed blocks [33]. Each block is identified by a Cryptographic Content Identifier (CCID) derived from its hash. Peers locate content providers through a Kademlia-style Distributed Hash Table (DHT) [26]. Kubo [31], the reference implementation of IPFS, exposes an HTTP gateway that allows conventional web clients to retrieve content without native IPFS support. A gateway request may be satisfied from a local block store, retrieved from a directly connected peer through Bitswap, or resolved through DHTbased provider discovery [3, 15]. Unlike traditional content delivery infrastructures, IPFS operates in an environment where content availability and retrieval paths can vary significantly over time. The same object may be replicated across different providers, served through different network paths, or become temporarily unavailable as peers join and leave the network. Measurements have shown that retrieval latency and throughput vary widely depending on content popularity, replication levels, provider availability, and peer churn [11, 12, 34, 37]. Adaptive Bitrate Streaming. Protocols, such as DASH [14], segments video content and encodes each segment at multiple bitrate representations. A Media Presentation Description (MPD) specifies the available representations, while clients dynamically select a bitrate for each segment during playback [13]. ABR algorithms differ primarily in the signals they use to estimate future network conditions. Rate-based approaches track recent throughput measurements, buffer-based approaches such as BOLA [35] and BBA [21] rely on playback buffer occupancy, control-theoretic approaches such as MPC [41] optimize decisions over a short prediction horizon, and learning-based approaches such as Pensieve [25] learn adaptation policies directly from observed network behavior. Early studies demonstrated that ABR clients can react slowly to bandwidth changes, oscillate between bitrate levels, and behave unfairly when competing for shared network resources [1, 22]. Subsequent surveys and modern ABR systems [5, 7] continue to balance three
Telescope proxy (stateless)
Bandwidth estimator 𝑇𝑐 , 𝑇𝑔 , 𝑇𝑛
DASH.js client
HTTP
MPD rewriter
ABR selector (throughput / statistics)
Request router (Fiber)
Two-layer cache (memory → file)
CID GET
Kubo IPFS cluster (3 nodes)
OpenTelemetry → Prometheus + Jaeger
Figure 1: Architecture of the re-engineered Telescope proxy. The stateless proxy mediates between the DASH.js client and a Kubo IPFS cluster, serving segments from a two-layer cache.
competing objectives: maximizing video quality, minimizing rebuffering events, and reducing quality oscillations. Achieving this balance relies on the assumption that recent network observations remain predictive of near-future delivery conditions. Network-Aware ABR. Several systems have proposed incorporating IPFS-specific network information into bitrate adaptation decisions. Telescope [40], the closest prior work, places a proxy between the IPFS gateway and the client, maintains throughput histories for both gateway and network retrieval paths, and rewrites the MPD so that the client can adapt to observed IPFS retrieval conditions. Experimental results demonstrate that exposing network-level retrieval information improves Quality of Experience (QoE), measured through a combination of average video quality, re-buffering events, and quality stability [16, 28]. More broadly, data-driven ABR systems such as CS2P [36] improve adaptation decisions through historical throughput observations and prediction models. Networkassisted adaptation has also been explored in other dynamic environments. For example, SARA [17] incorporates predictions of connectivity disruptions in LEO satellite networks into bitrate adaptation decisions to reduce re-buffering and quality degradation. These systems demonstrate that exposing additional network information can improve bitrate selection under challenging network conditions. Existing IPFS-aware ABR systems improve adaptation by maintaining historical information about network and provider behavior.
3
Stateless Network-Aware ABR Design
Our design pursues a different approach. Each bitrate decision is derived from observations available at request time rather than from long-lived adaptation state. The resulting design preserves network-awareness while avoiding synchronized throughput histories, provider-performance tables, or other shared adaptation metadata. Figure 1 illustrates the proposed architecture. As shown in the Figure 1, the system is deployed as a transparent proxy between a streaming client and a IPFS gateway. For each video request, the proxy observes current retrieval conditions, estimates
Stateless Network-Aware Adaptive Bitrate Streaming over IPFS
the bandwidth available to the client, and rewrites the MPD to expose only representations that can be sustained under those conditions. The client then continues to use its native ABR algorithm on the modified representation ladder. Because each decision is made independently from current observations, any proxy replica can serve any request without coordinating adaptation state with other replicas. The proxy estimates available bandwidth using three runtime signals: (i) client throughput, 𝑇𝑐 , measured as the recent segment delivery rate observed by the player, (ii) gateway fetch latency, 𝑇𝑔 , representing the time required to retrieve content from the IPFS gateway, and (iii) IPFS node throughput, 𝑇𝑛 , capturing the bandwidth of recent network retrievals from the IPFS network. The proxy additionally observes whether the requested segment is available in the local cache. Using these signals, the proxy derives an estimate of the bandwidth available for video delivery and removes bitrate representations that cannot be sustained under current conditions. This allows the client ABR algorithm to begin adaptation from a realistic representation set rather than discovering bandwidth limitations through playback stalls. We implement two alternative adaptation strategies. The throughputbased strategy estimates the usable video bandwidth by discounting the IPFS-related retrieval overhead from the observed client throughput: Bandwidth = 𝑇𝑐 − 𝑇𝑔
(cache hit)
(1)
Bandwidth = 𝑇𝑐 − 𝑇𝑛
(cache miss)
(2)
The statistics-based strategy combines cached and uncached bandwidth estimates to account for the possibility that future segment requests may encounter different cache states: Bandwidth = 31 cached + 23 uncached
(cache hit)
(3)
Bandwidth = 13 uncached + 23 cached
(cache miss)
(4)
The weighting factors are inherited from the reference implementation and are not tuned in this work. Since IPFS retrieval latency can significantly affect playback performance, each proxy maintains a two-tier cache consisting of an in-memory cache and a file-backed cache. Requests are resolved from memory first, then disk, and finally from the IPFS network. The cache serves purely as a retrieval optimization and is independent of the adaptation logic. Consequently, cache contents may differ across replicas without affecting the correctness of adaptation decisions. The prototype is implemented as a lightweight Go service positioned between the DASH client and the Kubo gateway1 . The proxy processes client requests, gateway interactions, cache lookups, and telemetry collection through independent concurrent execution pipelines implemented using Go’s native concurrency primitives. Efficient communication between these components is performed through Go channels, which provide lightweight synchronization and message passing mechanisms [2]. To support evaluation and debugging, the system exports telemetry through OpenTelemetry, Prometheus, and Jaeger [8–10].
1 https://github.com/iliyami/telescope
4
Experimental Methodology
Testbed. We deploy a three-node Kubo [31] IPFS cluster with Docker Compose; all services run as containers with no network emulation. The content is the Big Buck Bunny reference video at four content sizes (5, 10, 20, and 30 MB), each encoded with ffmpeg [18] into a five-rung bitrate ladder (426p–4K at 1, 3, 6, 9, and 12 Mbps) and segmented into more than 200 chunks; a bootstrap service uploads the segments to IPFS and records their CIDs, and a DASH.js client streams through the proxy. QoE metric. Following the original formulation [40], we score QoE per playback as 𝑁
QoE =
1 ∑︁ 𝑞𝑛 − 𝜇 StallRate − QualityVariation, 𝑁 𝑛=1
(5)
where 𝑞𝑛 is the bitrate (Mbps) of segment 𝑛 over the 𝑁 segments of a playback, the stall rate is (total playback time − video length)/video length, the quality variation is the mean absolute bitrate change between consecutive segments, and 𝜇 = 6. QoE is thus in Mbpsequivalent units bounded above by the top ladder rate (12 Mbps), and we use it to compare configurations within the study. We sweep a full factorial of two adaptation strategies (throughputand statistics-based), three caching policies (none, in-memory, and file-based), and three proxy replication levels (one to three replicas), giving 18 configurations; each is run 5–10 times and the reported values are averages. Network-level quantities (IPFS bandwidth, round-trip time, and hop count) are observed per run rather than controlled, so any relationship we report between them and QoE is correlational. We organize the evaluation around four questions: (RQ1) how much, and through which tier, does caching affect QoE and stalls; (RQ2) does throughput- or statistics-based adaptation perform better; (RQ3) does proxy replication improve QoE, and finally (RQ4) using the existing metrics, can we predict the QoE for decisions?
5
Results
We report results per research question. All values are read from the measured figures, and we compare configurations within the study rather than against an external baseline. Across all experiments, a consistent pattern emerges: clientperceived quality is governed primarily by content availability rather than by network topology or proxy deployment structure. Configurations that increase cache effectiveness consistently reduce stalls and improve QoE, while changes to adaptation strategy and replication level have comparatively smaller effects. The results therefore suggest a hierarchy of influence on streaming performance: caching is the dominant factor, adaptation strategy provides secondary gains when cache hits are unavailable, and replication mainly improves resilience and fetch-path diversity rather than user-visible quality. This observation is significant for stateless adaptation because the strongest predictors of QoE—recent throughput and cache availability—are both observable at request time and require no synchronized adaptation state across clients, providers, or proxy replicas.
Iliya Mirzaei, Shabnam Jafarzade Mojaveri, and Amirhossein Najafizadeh
Figure 2: QoE versus average stall rate by caching strategy.
Figure 4: Average IPFS bandwidth per adaptation strategy.
Figure 3: QoE versus IPFS bandwidth across adaptation strategies. Figure 5: Average QoE by proxy replication level.
5.1
RQ1: Caching
5.3
RQ3: Replication
Caching is the dominant QoE lever. Figure 2 shows QoE rising sharply as the cache absorbs retrieval latency: in-memory caching clusters at low stall rate (≈ 0.05–0.10) and high QoE (≈ 10.7–11.6), file-based caching sits close behind (≈ 9.2–10.2), and no-caching spreads to high stall rates (≈ 0.65–1.45) and low QoE (≈ 1.8–8.4). The gap widens with content size: for the largest clip, QoE falls from ≈ 11.6 with in-memory caching to 1.80 with none, a roughly 6× difference, and telemetry recorded cache hit rates rising by up to ≈ 60% under layered caching. In-memory yields the best QoE but is bounded by RAM, so file-based caching is the practical default, trading a small QoE reduction for durability and capacity.
Replication does not raise QoE. Figure 5 shows average QoE decreasing slightly from ≈ 8.9 at one replica to ≈ 8.1 at three, a change on the order of the run-to-run variability. Because each stateless replica holds an independent cache, spreading requests across replicas lowers per-replica hit rates; we did not measure per-replica hit rates directly, so we offer cache dilution as a conjecture rather than a confirmed mechanism. Replication still reduced average fetch latency in congested, high-hop runs (by up to ≈ 45% in telemetry), so it aids resilience even when QoE gains are marginal. For QoEoriented scale-out a shared caching tier is therefore preferable to more independent replicas.
5.2
5.4
RQ2: Adaptation strategy
Throughput-based adaptation is the more robust default. Figure 3 shows QoE rising with available IPFS bandwidth and saturating around 10–12 for both strategies, but the throughput-based strategy makes fuller use of capacity: it attains higher average IPFS bandwidth (≈ 24.5 versus ≈ 18 Mbps, Fig. 4) and higher gateway bandwidth (≈ 540 versus ≈ 447 Mbps). Its advantage is largest without caching, where raw retrieval latency dominates; once a cache absorbs that latency the two strategies converge (in-memory QoE ≈ 11, file-based ≈ 9–10 for both).
RQ4: What predicts QoE
Figure 6 summarises pairwise correlations among the measured metrics. Stall rate is almost perfectly inversely correlated with QoE (𝑟 = −0.98) and bitrate positively so (𝑟 = +0.51); both are partly definitional, as they are terms of Eq. (5). The informative result is that cache ratio, which is not a term of the metric, is strongly correlated with QoE (𝑟 = +0.73), whereas proxy replication (𝑟 = −0.11) and IPFS hop count (𝑟 ≈ 0) are essentially uncorrelated. Caching, not topology, governs client-perceived quality, reinforcing RQ1 and RQ3.
Stateless Network-Aware Adaptive Bitrate Streaming over IPFS
Figure 6: Correlation heatmap of the measured numerical metrics.
6
Discussion
Conclusion. This paper argues that maintaining synchronized adaptation state is a poor fit for network-aware ABR over IPFS, where provider churn, content redistribution, and changing retrieval paths can quickly invalidate stored information. We presented a soft-stateless design that recomputes bitrate decisions from request-time observations alone, requiring no shared adaptation state across clients, providers, or proxy replicas. Experimental results on a real Kubo testbed show that the design sustains high QoE, achieving up to roughly 6× the QoE of a non-adaptive, uncached configuration. Across all evaluated configurations, caching proved to be the dominant factor affecting QoE, throughput-based adaptation emerged as the most robust default strategy, and proxy replication improved resilience but did not improve QoE. These findings suggest that stateless adaptation is a practical foundation for decentralized video delivery systems. In this paper we characterize configurations of our system relative to one another and includes no external baseline, neither the original Telescope nor a non-IPFS-aware client, so we make no claim of improvement over prior systems. More fundamentally, the robustness and scalability benefits we attribute to statelessness are architectural arguments, not measured outcomes. Taken together, the results indicate that QoE in IPFS-based streaming is driven more by retrieval efficiency than by network structure. Caching consistently improves content availability and reduces stalls, throughput-based adaptation extracts additional performance when retrieval latency
is exposed, and replication primarily improves resilience rather than viewing quality. The strong correlation between cache ratio and QoE, combined with the weak correlation of replication and hop count, suggests that maintaining complex distributed adaptation state provides limited practical benefit. Instead, high-quality streaming can be achieved largely through local observations and efficient content reuse, supporting the feasibility of stateless network-aware ABR in decentralized environments.
Future Work. Several promising directions remain for future work. A direct comparison with stateful network-aware ABR systems, such as Telescope, would help quantify the resilience and scalability advantages of stateless adaptation under provider churn, failures, network partitions, and large-scale deployments. Extending the evaluation to larger and more geographically distributed IPFS environments would further validate the generality of the approach. Future research can also explore stateless adaptation policies specifically designed for request-time observations, including lightweight prediction and learning-based techniques. Finally, our results identify cache dilution as the primary limitation of independent replica scaling, motivating shared-content caching architectures that preserve stateless operation while improving cache efficiency and overall system performance. Collectively, these directions offer a path toward scalable, fault-tolerant, and fully decentralized video streaming systems built on IPFS.
Iliya Mirzaei, Shabnam Jafarzade Mojaveri, and Amirhossein Najafizadeh
References [1] Saamer Akhshabi, Ali C. Begen, and Constantine Dovrolis. 2011. An Experimental Evaluation of Rate-Adaptation Algorithms in Adaptive Streaming over HTTP. In Proceedings of the 2nd Annual ACM Conference on Multimedia Systems. doi:10. 1145/1943552.1943572 [2] Amirhossein Najafizadeh and Seyyed Ahmad Javadi. 2023. Jump over Golang channels. ResearchGate (2023). doi:10.13140/RG.2.2.28045.92646 [3] Leonhard Balduf, Sebastian Henningsen, Martin Florian, Sebastian Rust, and Björn Scheuermann. 2022. Monitoring Data Requests in Decentralized Data Storage Systems: A Case Study of IPFS. arXiv preprint arXiv:2104.09202 (2022). [4] Juan Benet. 2014. IPFS – Content Addressed, Versioned, P2P File System. arXiv:1407.3561. [5] Abdelhak Bentaleb, Bayan Taani, Ali C. Begen, Christian Timmerer, and Roger Zimmermann. 2019. A Survey on Bitrate Adaptation Schemes for Streaming Media Over HTTP. IEEE Communications Surveys & Tutorials (2019). [6] Divyashri Bhat, Amr Rizk, Michael Zink, and Ralf Steinmetz. 2018. SABR: Network-assisted content distribution for QoE-driven ABR video streaming. ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM) 14, 2s (2018), 1–25. [7] Tianyu Chen, Yiheng Lin, Nicolas Christianson, Zahid Akhtar, Sharath Dharmaji, Mohammad Hajiaghayi, Adam Wierman, and Ramesh K. Sitaraman. 2024. SODA: An Adaptive Bitrate Controller for Consistent High-Quality Video Streaming. In Proceedings of ACM SIGCOMM. [8] Cloud Native Computing Foundation. 2026. Jaeger: Open source, end-to-end distributed tracing. https://www.jaegertracing.io. Accessed: 2026-06-14. [9] Cloud Native Computing Foundation. 2026. OpenTelemetry: An observability framework for cloud-native software. https://opentelemetry.io. Accessed: 202606-14. [10] Cloud Native Computing Foundation. 2026. Prometheus: Monitoring system and time series database. https://prometheus.io. Accessed: 2026-06-14. [11] Erik Daniel and Florian Tschorsch. 2022. IPFS and Friends: A Qualitative Comparison of Next Generation Peer-to-Peer Data Networks. IEEE Communications Surveys & Tutorials (2022). [12] Erik Daniel and Florian Tschorsch. 2022. Passively Measuring IPFS Churn and Network Size. arXiv preprint arXiv:2205.14927 (2022). [13] DASH Industry Forum. 2026. dash.js: A reference client implementation for MPEG-DASH. https://github.com/Dash-Industry-Forum/dash.js. Accessed: 202606-14. [14] DASH Industry Forum. 2026. MPEG-DASH: Dynamic Adaptive Streaming over HTTP. https://dashif.org. Accessed: 2026-06-14. [15] Alfonso de la Rocha, David Dias, and Yiannis Psaras. 2021. Accelerating Content Routing with Bitswap: A Multi-Path File Transfer Protocol in IPFS and Filecoin. Technical Report. Protocol Labs Research. [16] Florin Dobrian, Vyas Sekar, Asad Awan, Ion Stoica, Dilip Joseph, Aditya Ganjam, Jibin Zhan, and Hui Zhang. 2011. Understanding the Impact of Video Quality on User Engagement. In ACM SIGCOMM 2011. [17] Hao Fang, Haoyuan Zhao, Jianxin Shi, Miao Zhang, Guanzhen Wu, Yi Ching Chou, Feng Wang, and Jiangchuan Liu. 2024. Robust Live Streaming over LEO Satellite Constellations: Measurement, Analysis, and Handover-Aware Adaptation. In Proceedings of the 32nd ACM International Conference on Multimedia. doi:10.1145/3664647.3680712 [18] FFmpeg Project. 2026. FFmpeg: A complete cross-platform solution to record, convert and stream audio and video. https://ffmpeg.org. Accessed: 2026-06-14. [19] Prateesh Goyal, Ravi Netravali, Mohammad Alizadeh, and Hari Balakrishnan. 2019. Secure incentivization for decentralized content delivery. In 2nd USENIX Workshop on Hot Topics in Edge Computing (HotEdge 19). [20] Jun He and Deying Yu. 2026. Post-Deterministic Distributed Systems: A New Foundation for Trustworthy Autonomous Infrastructure. arXiv preprint arXiv:2606.01722 (2026). [21] Te-Yuan Huang, Ramesh Johari, Nick McKeown, Matthew Trunnell, and Mark Watson. 2014. A Buffer-Based Approach to Rate Adaptation: Evidence from a Large Video Streaming Service. In ACM SIGCOMM 2014. [22] Junchen Jiang, Vyas Sekar, and Hui Zhang. 2012. Improving Fairness, Efficiency, and Stability in HTTP-Based Adaptive Video Streaming with FESTIVE. In Proceedings of the 8th International Conference on Emerging Networking Experiments and Technologies. doi:10.1145/2413176.2413189 [23] Amirhossein Khanlari and Amir Rahmati. 2026. Taking a Bite Out of the Forbidden Fruit: Characterizing Third-Party Iranian iOS App Stores. arXiv preprint arXiv:2604.26343 (2026). [24] Kyoungmin Kim, Youngin You, Mookyu Park, and Kyungho Lee. 2018. DDoS mitigation: Decentralized CDN using private blockchain. In 2018 Tenth International Conference on Ubiquitous and Future Networks (ICUFN). IEEE, 693–696. [25] Hongzi Mao, Ravi Netravali, and Mohammad Alizadeh. 2017. Neural Adaptive Video Streaming with Pensieve. In ACM SIGCOMM 2017. [26] Petar Maymounkov and David Mazières. 2002. Kademlia: A Peer-to-Peer Information System Based on the XOR Metric. In Proc. International Workshop on Peer-to-Peer Systems (IPTPS).
[27] Shota Minegishi and Masaki Bandai. 2025. Video segment prefetching for IPFS video streaming with scalable video coding. IEICE Transactions on Communications (2025). [28] Ricky K. P. Mok, Edmond W. W. Chan, and Rocky K. C. Chang. 2011. Measuring the Quality of Experience of HTTP Video Streaming. In IFIP/IEEE International Symposium on Integrated Network Management (IM). [29] Basabdatta Palit, Argha Sen, Abhijit Mondal, Ayan Zunaid, Jay Jayatheerthan, and Sandip Chakraborty. 2023. Improving UE energy efficiency through networkaware video streaming over 5G. IEEE Transactions on Network and Service Management 20, 3 (2023), 3487–3500. [30] Protocol Labs. 2026. IPFS: The InterPlanetary File System. https://ipfs.tech. Accessed: 2026-06-14. [31] Protocol Labs. 2026. Kubo: IPFS implementation in Go. https://github.com/ipfs/ kubo. Accessed: 2026-06-14. [32] Chunyu Qiao, Tong Liu, Yucheng Zhang, Zhiwei Fan, Pengjin Xie, Zhen Wang, and Liang Liu. 2025. PIRA: Pan-CDN Intra-video Resource Adaptation for Short Video Streaming. In Proceedings of the 33rd ACM International Conference on Multimedia. 11987–11995. [33] Hector Sanjuan, Samuli Poyhtari, Pedro Teixeira, and Ioannis Psaras. 2020. Merkle-crdts: Merkle-dags meet crdts. arXiv preprint arXiv:2004.00107 (2020). [34] Ruizhe Shi, Ruizhi Cheng, Bo Han, Yue Cheng, and Songqing Chen. 2024. A Closer Look into IPFS: Accessibility, Content, and Performance. Proceedings of the ACM on Measurement and Analysis of Computing Systems 8, 2, Article 20 (2024). doi:10.1145/3656015 [35] Kevin Spiteri, Rahul Urgaonkar, and Ramesh K. Sitaraman. 2016. BOLA: NearOptimal Bitrate Adaptation for Online Videos. In IEEE INFOCOM 2016. [36] Yi Sun, Xiaoqi Yin, Junchen Jiang, Vyas Sekar, Fuyuan Lin, Nanshu Wang, Tao Liu, and Bruno Sinopoli. 2016. CS2P: Improving Video Bitrate Selection and Adaptation with Data-Driven Throughput Prediction. In Proceedings of ACM SIGCOMM. doi:10.1145/2934872.2934898 [37] Dennis Trautwein, Aravindh Raman, Gareth Tyson, Ignacio Castro, Will Scott, Moritz Schubotz, Bela Gipp, and Yiannis Psaras. 2022. Design and Evaluation of IPFS: A Storage Layer for the Decentralized Web. In ACM SIGCOMM 2022. [38] Anuj Tyagi. 2025. Optimizing digital experiences with content delivery networks: Architectures, performance strategies, and future trends. arXiv preprint arXiv:2501.06428 (2025). [39] Thang X Vu, Symeon Chatzinotas, and Björn Ottersten. 2019. Blockchain-based content delivery networks: Content transparency meets user privacy. In 2019 IEEE Wireless Communications and Networking Conference (WCNC). IEEE, 1–6. [40] Zhengyu Wu, ChengHao Ryan Yang, Santiago Vargas, and Aruna Balasubramanian. 2023. Is IPFS Ready for Decentralized Video Streaming?. In Proceedings of the ACM Web Conference 2023 (WWW ’23). ACM. doi:10.1145/3543507.3583404 [41] Xiaoqi Yin, Abhishek Jindal, Vyas Sekar, and Bruno Sinopoli. 2015. A ControlTheoretic Approach for Dynamic Adaptive Video Streaming over HTTP. In ACM SIGCOMM 2015.