A Beginners Guide to Building a RAG App Using Open Source Milvus
Delay Analysis of Layered Video Caching in Crowdsourced Heterogeneous Wireless Networks
1. Delay Analysis of Layered Video Caching
in Crowdsourced Heterogeneous Wireless Networks
Behrouz Jedari and Mario Di Francesco
Department of Computer Science, Aalto University, Finland
Email: {behrouz.jedari,mario.di.francesco}@aalto.fi
11 December 2018
The 37th IEEE Global Communications Conference
(GLOBECOM 2018)
2. Introduction
• Exponential growth of mobile data traffic [Ericsson, 2018]
• Video data accounts for over 70% of the traffic
• New spectrum or multi-antenna techniques costly and
time-consuming
• Caching at the network edge is cost-efficient and agile
• User-provided networks enable crowdsourced caching
2
Ericsson, “Ericsson mobility report” June 2018, Available: https://www.ericsson.com/en/mobility-report/reports/june-2018
Behrouz Jedari and Mario Di Francesco - Delay Analysis of Layered Video Caching
3. Edge Caching in Heterogenous Networks
(HetNets)
• Caching content in small-cell base stations (SCBSs)
• Additional capacity of user equipment (UEs)
• Bringing content as close as possible to UEs
• Improving the quality of experience (QoE) of UEs, while
reducing the network backhaul traffic
3Behrouz Jedari and Mario Di Francesco - Delay Analysis of Layered Video Caching
4. • A video in multiple qualities
• An adaptive bitrate algorithm
Video encoding models [Cisco Webcasts, 2014]
1.Segment-based:
A video consists of multiple
segments, where each
segment has multiple qualities
• Not flexible
• No encoding overhead
2. Layered or scalable video
coding (SVC):
A video has a basic layer and
multiple enhancement layers
• Very flexible
• Encoding overhead
Cisco Webcasts, “Emerging video technologies: H.265, SVC, and webRTC” 2014. Available: https://www.ciscolive.com
Video Encoding
4Behrouz Jedari and Mario Di Francesco - Delay Analysis of Layered Video Caching
5. • Extension of H.264/MPEG-4 AVC
• Temporal, spatial, and SNR scalability
• Support of ultra high-definition videos
• Dependency between layers in caching and delivery
• Drawback: ~10% overhead per enhancement layer
Layered or SVC Videos
5Behrouz Jedari and Mario Di Francesco - Delay Analysis of Layered Video Caching
6. • Multi-bitrate video caching
[Li et al., 2017]
• The importance of transmission reliability between UEs
[Chen et al., 2017]
• Shortcoming: the main focus on segment-based video
encoding
[Li et al., 2018]
W. Li, et al., ”On the performance of adaptive video caching over information-centric networks” IEEE ICC, pages 1–6, 2017.
Z. Chen, et al., ”Probabilistic caching in wireless D2D networks: Cache hit optimal versus throughput optimal” IEEE
Communications Letters, vol. 21. no. 3, pp. 584–587, 2017.
L. Li, et al., “A survey of caching techniques in cellular networks: Research issues and challenges in content placement and
delivery strategies” IEEE Communications Surveys Tutorials, vol. 20, no. 3, pp. 1710-1732, 2018.
Related Work
6Behrouz Jedari and Mario Di Francesco - Delay Analysis of Layered Video Caching
7. • Energy-efficient LVC: minimizing the energy
consumption of base stations
[Xie et al., 2017]
• LVC in multi-operator networks: reducing the delivery
delay up to 25% by cooperative caching and delivery
[Poularakis et al., 2016]
• LVC in crowdsourced HetNets is overlooked!
J. Xie, et al.,“Energy-efficient content placement for layered video content delivery over cellular networks” IEEE GLOBECOM, 2017,
pp. 1–6.
K. Poularakis et al., “Caching and operator cooperation policies for layered video content delivery” IEEE INFOCOM, 2016, pp. 1–9.
Related Work (cont.)
7Behrouz Jedari and Mario Di Francesco - Delay Analysis of Layered Video Caching
8. 1.How to maximize the QoE of UEs in LVC?
2.Which video layers should be cached, so that the
average download time is minimized?
8
Research Questions
Behrouz Jedari and Mario Di Francesco - Delay Analysis of Layered Video Caching
9. • System Model
• Problem Formulation
• Proposed Solution
• Numerical Results
• Conclusion
Outline
9Behrouz Jedari and Mario Di Francesco - Delay Analysis of Layered Video Caching
10. Multiple SCBSs in the coverage of one macro BS (MBS)
10
Network Model
Four scenarios in
content delivery:
a) Self-response
b) UPN-response
c) SCBS-response
d) MBS-response
Behrouz Jedari and Mario Di Francesco - Delay Analysis of Layered Video Caching
11. •Videos with skewed popularity
•Zipf-like distribution to model the video popularity (i.e., the
request probability) [Cha et al., 2007]
• The popularity of i-th ranked video is calculated as:
where ! denotes the popularity skewness
M. Cha, et al., “I tube, you tube, everybody tubes: Analyzing the world’s largest user generated content video system” ACM
SIGCOMM conference on Internet measurement, 2007, pp. 1–14.
11
User Demand Model
Behrouz Jedari and Mario Di Francesco - Delay Analysis of Layered Video Caching
12. Videos are sorted based on their popularity
Video Caching Model
12Behrouz Jedari and Mario Di Francesco - Delay Analysis of Layered Video Caching
13. Problem: delay minimization in LVC and delivery
Objective: maximizing the total delay saving
Total delay
without caching
Total delay with
caching in both the
SCBSs and UEs
max
13
Problem Formulation
Behrouz Jedari and Mario Di Francesco - Delay Analysis of Layered Video Caching
14. Constraints
Caching capacity in
SCBSs and UEs
Dependencies
between video layers
Problem Formulation (cont.)
14Behrouz Jedari and Mario Di Francesco - Delay Analysis of Layered Video Caching
15. Problem: delay minimization in LVC
Objective: maximizing the total delay saving
Total delay without
caching
Total delay with caching
in SCBSs and UEs
max
15
Problem Formulation (cont.)
An instance of fractional knapsack problem (non-
convex and NP-hard) [Ishii et al., 1977]
H. Ishii, et al., “Fractional knapsack problems” Mathematical Programming, vol. 13, no. 1, pp. 255–271, 1977.
Behrouz Jedari and Mario Di Francesco - Delay Analysis of Layered Video Caching
16. • Difference of convex (DC) programming: a near-optimal
solution [H. A. Le Thi et al., 2014]
• Idea: the combination of two convex functions is convex
• Decompose the objective function into two convex functions
where
H. A. Le Thi, et al., “DC programming and DCA for general DC programs” Advanced Computational Methods for Knowledge
Engineering, 2014, pp. 15–35.
16
Proposed Solution
Behrouz Jedari and Mario Di Francesco - Delay Analysis of Layered Video Caching
17. Assign initial values
Optimize the ILP
problem iteratively
17
DC Programming for the LVC problem
Behrouz Jedari and Mario Di Francesco - Delay Analysis of Layered Video Caching
18. • The evaluation with T=1000 time slots
• One MBS, four SCBSs, and 300 UEs
• Videos generation by [Video trace library] (each video has five layers)
• SCBSs and UEs according to Poisson point process (PPP)
Four scenarios
“Video trace library,” http://trace.eas.asu.edu/.
1.NoCache
2.LVC-NoUPN
3.RandomCache
4.LVC-Optimized
Simulation Setup
18Behrouz Jedari and Mario Di Francesco - Delay Analysis of Layered Video Caching
19. Obs. 1: LVC-Optimized achieves the lowest delay
Obs. 2: LVC-NoUPN outperforms RandomCache
19
Impact of Cache Size
Behrouz Jedari and Mario Di Francesco - Delay Analysis of Layered Video Caching
20. Obs. 1: the delay in LVC-Optimized grows “slowly”
Obs. 2: LVC-NoUPN and RandomCache have similar performance
20
Impact of User Demand and Video Popularity
Behrouz Jedari and Mario Di Francesco - Delay Analysis of Layered Video Caching
21. — Significant improvement of cache hit ratio by caching
contents in both the SCBSs and UEs
— Optimal LVC is an NP-hard problem
— DC programming to find a near-optimal solution
— Several open problems in LVC
21
Conclusion
Behrouz Jedari and Mario Di Francesco - Delay Analysis of Layered Video Caching
22. Thanks for your attention
Behrouz Jedari
Postdoctoral researcher at Aalto University, Finland
Home page: https://sites.google.com/site/behrouzjedari
Email: behrouz.jedari@aalto.fi
The 37th IEEE Global Communications Conference
(GLOBECOM 2018)