Source-linked AI summary
Understanding Performance of Edge Content Caching for Mobile Video Streaming
Ge Ma, Zhi Wang, Miao Zhang, Jiahui Ye, Minghua Chen, Wenwu Zhu
TL;DR
The paper addresses limited understanding of mobile-video request patterns and edge-caching effectiveness in practical wireless networks. Using large-scale real-world traces, it analyzes request patterns and user behavior, compares Wi-Fi and cellular caching strategies, and designs a geo-collaborative strategy with reported improvements over LRU and LFU.
Problem
Mobile-video request patterns and the effectiveness of Wi-Fi and cellular edge caching are unclear in practical wireless networks.
Method
The study analyzes two weeks of large-scale mobile-video traces using frequency-domain and entropy analyses, mobility analysis, and trace-driven comparisons of LRU, LFU, and a geo-collaborative strategy.
Results
The geo-collaborative strategy achieves 20% cache-hit-rate and service-rate improvements over LRU and 30% improvements over LFU.
Takeaways & Limitations
Caching performance depends on request patterns, user mobility, content popularity, cache capacity, caching strategy, and wireless-network deployment.
Abstract
from arXiv · showhide
Today's Internet has witnessed an increase in the popularity of mobile video streaming, which is expected to exceed 3/4 of the global mobile data traffic by 2019. To satisfy the considerable amount of mobile video requests, video service providers have been pushing their content delivery infrastructure to edge networks--from regional CDN servers to peer CDN servers (e.g., smartrouters in users' homes)--to cache content and serve users with storage and network resources nearby. Among the edge network content caching paradigms, Wi-Fi access point caching and cellular base station caching have become two mainstream solutions. Thus, understanding the effectiveness and performance of these solutions for large-scale mobile video delivery is important. However, the characteristics and request patterns of mobile video streaming are unclear in practical wireless network. In this paper, we use real-world datasets containing 50 million trace items of nearly 2 million users viewing more than 0.3 million unique videos using mobile devices in a metropolis in China over 2 weeks, not only to understand the request patterns and user behaviors in mobile video streaming, but also to evaluate the effectiveness of Wi-Fi and cellular-based edge content caching solutions. To understand performance of edge content caching for mobile video streaming, we first present temporal and spatial video request patterns, and we analyze their impacts on caching performance using frequency-domain and entropy analysis approaches. We then study the behaviors of mobile video users, including their mobility and geographical migration behaviors. Using trace-driven experiments, we compare strategies for edge content caching including LRU and LFU, in terms of supporting mobile video requests. Moreover, we design an efficient caching strategy based on the measurement insights and experimentally evaluate its performance.
I. INTRODUCTION
The paper examines mobile video request patterns, user behavior, and edge-network caching using large-scale measurements, then designs a geo-collaborative caching strategy. It compares Wi-Fi and cellular caching and reports improvements over LRU and LFU.
- The study jointly considers mobile video service and wireless-network perspectives to analyze Wi-Fi and cellular infrastructures and improve wireless-network QoS according to request patterns.
- 50 million sessions from nearly 2 million users viewing more than 0.3 million videos over two weeks support the study of real-world mobile video behavior.
- Mobile video requests exhibit distinctive spatial and temporal patterns that can significantly affect edge-content caching performance.
- Users show heterogeneous mobility and regular geographic migration, including two to three regularly visited locations and repeated movement between similar location types.
- The authors design a geo-collaborative caching strategy that jointly considers request patterns, user behaviors, and wireless-network deployment.
- 20% and 30% cache-hit-rate improvements, respectively, are reported over conventional LRU and LFU caching strategies.The same comparison reports 20% and 30% service-rate improvements, respectively.
B. User Mobility Behaviors
This material situates mobile video caching within edge-delivery research and describes the datasets and measurement basis used to study user mobility and wireless-network deployment.
- Mobile video caching addresses challenges to traditional CDN paradigms created by substantial bandwidth demand from data-intensive applications.
- The paper claims novelty from jointly measuring mobile-video requests, user mobility, and wireless-network deployment to study caching performance and design a caching strategy.
- The study investigates mobile video behavior using a dataset spanning two weeks and covering about two million users and 0.3 million videos in Beijing.
- The behavior dataset records device identifiers, viewing timestamps, viewing locations, and video titles for mobile-video sessions.
B. Wi-Fi and Cellular Network Dataset
The paper examines Wi-Fi and cellular edge-network datasets alongside mobile video popularity and request geography to assess how deployment and locality affect caching. It identifies local variation in video popularity and argues that delivery should account for geographic demand and user movement.
- The study considers both Wi-Fi and cellular solutions for supporting edge-network content delivery.
- The Wi-Fi dataset covers over 1 million Beijing access points, while the cellular dataset contains information on over 70 thousand base stations.
- Mobile video popularity follows a power-law distribution, with category-specific differences in request decay and content lifetime.Trailers decrease most slowly, short variety shows decrease fastest, and animation has the longest lifetime.
- The top 0.3% of videos have substantially different popularity ranks across locations, so global popularity does not directly determine local popularity.Their average rank is within the top 40% in 60% of locations.
- Users are classified as multi-location or single-location according to whether their requests within one day come from different or the same access points or base stations.
- Edge delivery should account for geographic request density, allocate more resources to high-demand locations, and proactively push content during off-peak times.
2) Multi-location Users in Different Locations:
The paper analyzes periodic request patterns with frequency-domain methods and studies geographic variation in video requests with entropy measures. These analyses connect recurring traffic structure and content-location diversity to edge caching decisions.
- Periodic video requests can affect edge caching strategies, including content replication and resource allocation.
- Frequency Analysis of Periodical Request Patterns: The frequency analysis represents one week of hourly request counts as a sequence with N = 168 time slots.
- Frequency Analysis of Periodical Request Patterns: The DFT produces a frequency spectrum in which larger X[k] indicates a stronger period of k.Amplitude represents request volume, while phase represents the timing of peaks and valleys.
- Frequency Analysis of Periodical Request Patterns: Major frequencies k = 7, 14, and 21 correspond to periods of 1 day, 12 hours, and 8 hours, respectively.These components can represent the time-domain traffic.
- Content Video: An Entropy Approach: The study uses entropy analysis to examine how different videos are requested across different locations, defining video-request and location-request entropy.
1) Geographical Video Request Entropy:
The paper uses geographical and location request entropy to characterize how mobile video requests are distributed across locations and videos. These patterns vary with video popularity, location functionality, and user mobility, with implications for caching strategy selection.
- Location request entropy measures video diversity within a location, and larger values generally require more content items to be replicated.
- The entropy comparisons require caution because adding locations increases entropy, making direct comparisons unfair when the numbers of locations differ.
- Geographical video request entropy increases with video popularity, indicating that popular videos receive requests from more locations.
- LFU is better suited to locations with smaller entropy, whereas LRU is better suited to locations with larger entropy.
- Locations with more PoI functionality labels typically have smaller request entropy and more diverse, skewed video-request distributions.
V. MOBILE VIDEO REQUESTS AFFECTED BY USER MOBILITY BEHAVIORS
The mobility analysis shows that active mobile video users exhibit varied but generally limited movement across locations. Consecutive-request distances and intervals depend on elapsed time and movement speed, shaping how requests reach edge infrastructure.
- Among 9,576 active users, 30% are multi-location users and 70% are simple-location users.
- 80% of multi-location users request videos from fewer than 4 locations, while 50% use only 2 locations in one day.
- Consecutive requests made within 10 minutes occur over much shorter distances than requests separated by longer intervals.
- At speeds below 5.6 km/h, 80% of request intervals occur within 1.5 hours.
B. Migration Patterns
Users’ migrations are concentrated among a small set of location transitions. The dominant patterns include repeated movement between particular locations and substantial migration between selected PoI functionality types.
- The seven most popular migration patterns account for 70% of all migrations between locations.
- Movement between two particular locations constitutes almost 50% of migrations.
- Users commonly migrate between locations with the same PoI type, while some specific functionality pairs also have large migration counts.
- The largest migration count occurs between residential and business areas.
A. Request Coverage by Edge Network Infrastructure
The paper evaluates whether Wi-Fi APs and cellular base stations are geographically positioned to serve mobile video requests. Although infrastructure broadly follows demand across the city, high-intensity areas show a substantial mismatch, with Wi-Fi APs generally closer to users.
- Over 95% of requests can find a Wi-Fi AP within 500 meters or a cellular BS within 750 meters.
- More than 80% of request locations have a cellular BS farther away than the nearest Wi-Fi AP.
- Citywide request intensity and AP/BS deployment have cosine similarity above 0.77, but similarity falls below 0.39 in the centralized 30% area containing over 80% of requests.
- The deployment-demand mismatch is especially pronounced in high-intensity locations, where APs and BSes may not satisfy requests adequately.
- The evaluation uses a trace-driven simulator that routes requests to the nearest Wi-Fi AP or cellular BS and measures cache hit rates.
1) Experimental Setup:
The experiments use two weeks of real request traces to compare conventional caching strategies across Wi-Fi and cellular edge networks. Cache performance varies with capacity, diversity, and infrastructure type, while request density has only a limited effect.
- Experimental Setup: The simulator uses equal cache capacity across APs and BSes, with default capacity 20 items, concurrency 20/100, and bandwidth 20S/100S for APs/BSes.
- Experimental Setup: LRU, LFU, and random replacement are compared as conventional caching strategies.
- Cache Capacity: For LRU to reach cache hit rates of 0.25, 0.42, and 0.61, average Wi-Fi AP capacities are 5, 20, and 50, versus 12, 49, and 118 for cellular BSes.
- Cache Capacity: LRU, LFU, and random replacement achieve similar cache hit rates particularly when cache capacity is large.
- Request Density: Cache hit rate decreases only slightly as request density increases, indicating relative insensitivity to request density.
- Video and User Diversity: Lower video and user diversity typically produce higher cache hit rates, while Wi-Fi caching can retain high hit rates at larger diversity levels.
5) Impact of User Mobility:
User mobility reduces edge-cache effectiveness because users request different videos across locations. The proposed geo-collaborative strategy allocates cache resources and references content across locations using mobility and request-popularity information.
- Impact of User Mobility: Multi-location users have lower cache hit rates than single-location users under both LRU and LFU.
- Impact of User Mobility: Most location pairs have Jaccard similarity below 0.4 for the videos requested across locations.
- Impact of User Mobility: When multi-location users can fetch content from their starting location, caching performance improves greatly and LFU outperforms LRU.
- Geo-Collaborative Caching: The geo-collaborative strategy divides each location’s cache between single-location and multi-location users.
- Geo-Collaborative Caching: It ranks locations for cross-location content reference using user migration information.
- Geo-Collaborative Caching: Content for single-location users is selected using historical video popularity with a category-dependent positive decay factor.
B. Performance Evaluation
The proposed geo-collaborative strategy is evaluated against LRU and LFU using the same trace-driven simulator, focusing on the most requested edge locations. Its advantage grows with cache capacity, including for service rate.
- Evaluation Scope: The evaluation considers only the top 10% most requested cellular BSes or Wi-Fi APs to ensure sufficient requests per location.
- Cache Hit Rate: Increasing cache capacity raises cache hit rates for LRU, LFU, and the geo-collaborative strategy.
- Cache Hit Rate: The geo-collaborative strategy’s gain over LRU and LFU increases faster at the beginning as cache capacity grows.
- Service Rate: The geo-collaborative strategy’s service-rate gain over LRU and LFU gradually increases with cache capacity.
VIII. CONCLUDING REMARKS
The paper combines measurement studies with trace-driven experiments to examine mobile video request patterns, user behaviors, and edge caching performance. Its findings indicate that geographic and temporal request diversity, user mobility, and preferences make joint caching strategies promising.
- The study uses measurements and trace-driven experiments to investigate edge content caching for mobile video delivery.It examines spatial and temporal request patterns alongside user behaviors.
- Geographic request distributions are highly diverse, with requested content varying across locations and periods.
- User mobility and preferences generally determine mobile video request patterns, with regular commuting behavior suggesting joint caching strategies.
- The paper compares cellular- and Wi-Fi-based edge network caching solutions for mobile video delivery.