Top Banner
Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba¸ stu˘ g , Mehdi Bennis ? , Engin Zeydan , Manhal Abdel Kader , Alper Karatepe , Ahmet Salih Er , and M´ erouane Debbah , Large Networks and Systems Group (LANEAS), CentraleSup´ elec, Universit´ e Paris-Saclay, Gif-sur-Yvette, France ? Centre for Wireless Communications, University of Oulu, Finland AveaLabs, Istanbul, Turkey Mathematical and Algorithmic Sciences Lab, Huawei France R&D, Paris, France [email protected], [email protected].fi, {engin.zeydan, alper.karatepe, ahmetsalih.er}@avea.com.tr, [email protected] [email protected] The ETSI Workshop “From Research to Standardization” May 10 2016, Sophia Antipolis, FRANCE. L A N E A S Large Networks and Systems Group
52

Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Jun 10, 2020

Download

Documents

dariahiddleston
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Big Data Meets Telcos:A Proactive CachingPerspectiveEjder Bastug�, Mehdi Bennis?, Engin Zeydan◦, Manhal Abdel Kader�, Alper

Karatepe◦, Ahmet Salih Er◦, and Merouane Debbah�,†

�Large Networks and Systems Group (LANEAS), CentraleSupelec, Universite Paris-Saclay, Gif-sur-Yvette, France?Centre for Wireless Communications, University of Oulu, Finland

◦AveaLabs, Istanbul, Turkey†Mathematical and Algorithmic Sciences Lab, Huawei France R&D, Paris, France

[email protected], [email protected],{engin.zeydan, alper.karatepe, ahmetsalih.er}@avea.com.tr,

[email protected] [email protected]

The ETSI Workshop “From Research to Standardization”May 10 2016, Sophia Antipolis, FRANCE.

L A N E A SLarge Networks and Systems Group

Page 2: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

IntroductionMotivation

I Mobile cellular networks are becoming increasingly complex [And+14]

I Classical deployment/optimization techniques and solutions (i.e., celldensification, acquiring more spectrum, etc.) are cost-ineffective and thusseen as stopgaps

I This calls for development of novel approaches that leverage recentadvances in storage/memory, context-awareness, edge/cloud computing,and falls into framework of big data [BBD14]

I The big data has its notorious 4V: velocity, voracity, volume and variety

Based on these motivations, we focus on

Caching at the edge + enable big data!

In particular, our contributions:

I Collect users’ mobile traffic data from a telecom operator

I Characterize content popularity and size distributions

I Exploit machine learning tools and investigate gains of caching in terms ofusers’ satisfaction and backhaul offloading

[And+14] J.G. Andrews et al. “What Will 5G Be?” In: IEEE Journal on Selected Areas in Communications 32.6 (June 2014),pp. 1065–1082

[BBD14] Ejder Bastug et al. “Living on the Edge: The role of Proactive Caching in 5G Wireless Networks”. In: IEEE CommunicationsMagazine 52.8 (Aug. 2014), pp. 82–89 2/19

Page 3: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

IntroductionMotivation

I Mobile cellular networks are becoming increasingly complex [And+14]

I Classical deployment/optimization techniques and solutions (i.e., celldensification, acquiring more spectrum, etc.) are cost-ineffective and thusseen as stopgaps

I This calls for development of novel approaches that leverage recentadvances in storage/memory, context-awareness, edge/cloud computing,and falls into framework of big data [BBD14]

I The big data has its notorious 4V: velocity, voracity, volume and variety

Based on these motivations, we focus on

Caching at the edge + enable big data!

In particular, our contributions:

I Collect users’ mobile traffic data from a telecom operator

I Characterize content popularity and size distributions

I Exploit machine learning tools and investigate gains of caching in terms ofusers’ satisfaction and backhaul offloading

[And+14] J.G. Andrews et al. “What Will 5G Be?” In: IEEE Journal on Selected Areas in Communications 32.6 (June 2014),pp. 1065–1082

[BBD14] Ejder Bastug et al. “Living on the Edge: The role of Proactive Caching in 5G Wireless Networks”. In: IEEE CommunicationsMagazine 52.8 (Aug. 2014), pp. 82–89 2/19

Page 4: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

IntroductionMotivation

I Mobile cellular networks are becoming increasingly complex [And+14]

I Classical deployment/optimization techniques and solutions (i.e., celldensification, acquiring more spectrum, etc.) are cost-ineffective and thusseen as stopgaps

I This calls for development of novel approaches that leverage recentadvances in storage/memory, context-awareness, edge/cloud computing,and falls into framework of big data [BBD14]

I The big data has its notorious 4V: velocity, voracity, volume and variety

Based on these motivations, we focus on

Caching at the edge + enable big data!

In particular, our contributions:

I Collect users’ mobile traffic data from a telecom operator

I Characterize content popularity and size distributions

I Exploit machine learning tools and investigate gains of caching in terms ofusers’ satisfaction and backhaul offloading

[And+14] J.G. Andrews et al. “What Will 5G Be?” In: IEEE Journal on Selected Areas in Communications 32.6 (June 2014),pp. 1065–1082

[BBD14] Ejder Bastug et al. “Living on the Edge: The role of Proactive Caching in 5G Wireless Networks”. In: IEEE CommunicationsMagazine 52.8 (Aug. 2014), pp. 82–89 2/19

Page 5: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

IntroductionMotivation

I Mobile cellular networks are becoming increasingly complex [And+14]

I Classical deployment/optimization techniques and solutions (i.e., celldensification, acquiring more spectrum, etc.) are cost-ineffective and thusseen as stopgaps

I This calls for development of novel approaches that leverage recentadvances in storage/memory, context-awareness, edge/cloud computing,and falls into framework of big data [BBD14]

I The big data has its notorious 4V: velocity, voracity, volume and variety

Based on these motivations, we focus on

Caching at the edge + enable big data!

In particular, our contributions:

I Collect users’ mobile traffic data from a telecom operator

I Characterize content popularity and size distributions

I Exploit machine learning tools and investigate gains of caching in terms ofusers’ satisfaction and backhaul offloading

[And+14] J.G. Andrews et al. “What Will 5G Be?” In: IEEE Journal on Selected Areas in Communications 32.6 (June 2014),pp. 1065–1082

[BBD14] Ejder Bastug et al. “Living on the Edge: The role of Proactive Caching in 5G Wireless Networks”. In: IEEE CommunicationsMagazine 52.8 (Aug. 2014), pp. 82–89 2/19

Page 6: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

IntroductionMotivation

I Mobile cellular networks are becoming increasingly complex [And+14]

I Classical deployment/optimization techniques and solutions (i.e., celldensification, acquiring more spectrum, etc.) are cost-ineffective and thusseen as stopgaps

I This calls for development of novel approaches that leverage recentadvances in storage/memory, context-awareness, edge/cloud computing,and falls into framework of big data [BBD14]

I The big data has its notorious 4V: velocity, voracity, volume and variety

Based on these motivations, we focus on

Caching at the edge + enable big data!

In particular, our contributions:

I Collect users’ mobile traffic data from a telecom operator

I Characterize content popularity and size distributions

I Exploit machine learning tools and investigate gains of caching in terms ofusers’ satisfaction and backhaul offloading

[And+14] J.G. Andrews et al. “What Will 5G Be?” In: IEEE Journal on Selected Areas in Communications 32.6 (June 2014),pp. 1065–1082

[BBD14] Ejder Bastug et al. “Living on the Edge: The role of Proactive Caching in 5G Wireless Networks”. In: IEEE CommunicationsMagazine 52.8 (Aug. 2014), pp. 82–89 2/19

Page 7: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

IntroductionMotivation

I Mobile cellular networks are becoming increasingly complex [And+14]

I Classical deployment/optimization techniques and solutions (i.e., celldensification, acquiring more spectrum, etc.) are cost-ineffective and thusseen as stopgaps

I This calls for development of novel approaches that leverage recentadvances in storage/memory, context-awareness, edge/cloud computing,and falls into framework of big data [BBD14]

I The big data has its notorious 4V: velocity, voracity, volume and variety

Based on these motivations, we focus on

Caching at the edge + enable big data!

In particular, our contributions:

I Collect users’ mobile traffic data from a telecom operator

I Characterize content popularity and size distributions

I Exploit machine learning tools and investigate gains of caching in terms ofusers’ satisfaction and backhaul offloading

[And+14] J.G. Andrews et al. “What Will 5G Be?” In: IEEE Journal on Selected Areas in Communications 32.6 (June 2014),pp. 1065–1082

[BBD14] Ejder Bastug et al. “Living on the Edge: The role of Proactive Caching in 5G Wireless Networks”. In: IEEE CommunicationsMagazine 52.8 (Aug. 2014), pp. 82–89 2/19

Page 8: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

IntroductionSome Relevant Works

I Scaling laws for caching in content delivery networks [GPT12]

I FemtoCaching architecture [Gol+13]

I LivingOnTheEdge: Edge caching and content popularity learningaspects [BBD14]

I Deployment aspects of cache-enabled single-tier networks [Bas+15]

I Coded caching gains in physical layer [ZE15]

I MDS and regenerating codes [BGL15][Ped+15]

[GPT12] Savvas Gitzenis et al. “Asymptotic Laws for Joint Content Replication and Delivery in Wireless Networks”. In: IEEETransactions on Information Theory [Online] arXiv:1201.3095 59.12 (Dec. 2012), pp. 2760–2776

[Gol+13] Negin Golrezaei et al. “Femtocaching and device-to-device collaboration: A new architecture for wireless video distribution”. In:IEEE Communications Magazine 51.4 (2013), pp. 142–149

[BBD14] Ejder Bastug et al. “Living on the Edge: The role of Proactive Caching in 5G Wireless Networks”. In: IEEE CommunicationsMagazine 52.8 (Aug. 2014), pp. 82–89

[Bas+15] Ejder Bastug et al. “Cache-enabled Small Cell Networks: Modeling and Tradeoffs”. In: EURASIP Journal on WirelessCommunications and Networking 1 (Feb. 2015), p. 41

[ZE15] Jingjing Zhang and Petros Elia. “Fundamental Limits of Cache-Aided Wireless BC: Interplay of Coded-Caching and CSITFeedback”. In: [Online] arXiv:1511.03961 (2015)

[BGL15] Valerio Bioglio et al. “Optimizing MDS Codes for Caching at the Edge”. In: [Online] arXiv:1508.05753 (2015)

[Ped+15] Jesper Pedersen et al. “Repair Scheduling in Wireless Distributed Storage with D2D Communication”. In: [Online]arXiv:1504.06231 (2015)

3/19

Page 9: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

IntroductionSome Relevant Works

I Scaling laws for caching in content delivery networks [GPT12]

I FemtoCaching architecture [Gol+13]

I LivingOnTheEdge: Edge caching and content popularity learningaspects [BBD14]

I Deployment aspects of cache-enabled single-tier networks [Bas+15]

I Coded caching gains in physical layer [ZE15]

I MDS and regenerating codes [BGL15][Ped+15]

[GPT12] Savvas Gitzenis et al. “Asymptotic Laws for Joint Content Replication and Delivery in Wireless Networks”. In: IEEETransactions on Information Theory [Online] arXiv:1201.3095 59.12 (Dec. 2012), pp. 2760–2776

[Gol+13] Negin Golrezaei et al. “Femtocaching and device-to-device collaboration: A new architecture for wireless video distribution”. In:IEEE Communications Magazine 51.4 (2013), pp. 142–149

[BBD14] Ejder Bastug et al. “Living on the Edge: The role of Proactive Caching in 5G Wireless Networks”. In: IEEE CommunicationsMagazine 52.8 (Aug. 2014), pp. 82–89

[Bas+15] Ejder Bastug et al. “Cache-enabled Small Cell Networks: Modeling and Tradeoffs”. In: EURASIP Journal on WirelessCommunications and Networking 1 (Feb. 2015), p. 41

[ZE15] Jingjing Zhang and Petros Elia. “Fundamental Limits of Cache-Aided Wireless BC: Interplay of Coded-Caching and CSITFeedback”. In: [Online] arXiv:1511.03961 (2015)

[BGL15] Valerio Bioglio et al. “Optimizing MDS Codes for Caching at the Edge”. In: [Online] arXiv:1508.05753 (2015)

[Ped+15] Jesper Pedersen et al. “Repair Scheduling in Wireless Distributed Storage with D2D Communication”. In: [Online]arXiv:1504.06231 (2015)

3/19

Page 10: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

IntroductionSome Relevant Works

I Scaling laws for caching in content delivery networks [GPT12]

I FemtoCaching architecture [Gol+13]

I LivingOnTheEdge: Edge caching and content popularity learningaspects [BBD14]

I Deployment aspects of cache-enabled single-tier networks [Bas+15]

I Coded caching gains in physical layer [ZE15]

I MDS and regenerating codes [BGL15][Ped+15]

[GPT12] Savvas Gitzenis et al. “Asymptotic Laws for Joint Content Replication and Delivery in Wireless Networks”. In: IEEETransactions on Information Theory [Online] arXiv:1201.3095 59.12 (Dec. 2012), pp. 2760–2776

[Gol+13] Negin Golrezaei et al. “Femtocaching and device-to-device collaboration: A new architecture for wireless video distribution”. In:IEEE Communications Magazine 51.4 (2013), pp. 142–149

[BBD14] Ejder Bastug et al. “Living on the Edge: The role of Proactive Caching in 5G Wireless Networks”. In: IEEE CommunicationsMagazine 52.8 (Aug. 2014), pp. 82–89

[Bas+15] Ejder Bastug et al. “Cache-enabled Small Cell Networks: Modeling and Tradeoffs”. In: EURASIP Journal on WirelessCommunications and Networking 1 (Feb. 2015), p. 41

[ZE15] Jingjing Zhang and Petros Elia. “Fundamental Limits of Cache-Aided Wireless BC: Interplay of Coded-Caching and CSITFeedback”. In: [Online] arXiv:1511.03961 (2015)

[BGL15] Valerio Bioglio et al. “Optimizing MDS Codes for Caching at the Edge”. In: [Online] arXiv:1508.05753 (2015)

[Ped+15] Jesper Pedersen et al. “Repair Scheduling in Wireless Distributed Storage with D2D Communication”. In: [Online]arXiv:1504.06231 (2015)

3/19

Page 11: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

IntroductionSome Relevant Works

I Scaling laws for caching in content delivery networks [GPT12]

I FemtoCaching architecture [Gol+13]

I LivingOnTheEdge: Edge caching and content popularity learningaspects [BBD14]

I Deployment aspects of cache-enabled single-tier networks [Bas+15]

I Coded caching gains in physical layer [ZE15]

I MDS and regenerating codes [BGL15][Ped+15]

[GPT12] Savvas Gitzenis et al. “Asymptotic Laws for Joint Content Replication and Delivery in Wireless Networks”. In: IEEETransactions on Information Theory [Online] arXiv:1201.3095 59.12 (Dec. 2012), pp. 2760–2776

[Gol+13] Negin Golrezaei et al. “Femtocaching and device-to-device collaboration: A new architecture for wireless video distribution”. In:IEEE Communications Magazine 51.4 (2013), pp. 142–149

[BBD14] Ejder Bastug et al. “Living on the Edge: The role of Proactive Caching in 5G Wireless Networks”. In: IEEE CommunicationsMagazine 52.8 (Aug. 2014), pp. 82–89

[Bas+15] Ejder Bastug et al. “Cache-enabled Small Cell Networks: Modeling and Tradeoffs”. In: EURASIP Journal on WirelessCommunications and Networking 1 (Feb. 2015), p. 41

[ZE15] Jingjing Zhang and Petros Elia. “Fundamental Limits of Cache-Aided Wireless BC: Interplay of Coded-Caching and CSITFeedback”. In: [Online] arXiv:1511.03961 (2015)

[BGL15] Valerio Bioglio et al. “Optimizing MDS Codes for Caching at the Edge”. In: [Online] arXiv:1508.05753 (2015)

[Ped+15] Jesper Pedersen et al. “Repair Scheduling in Wireless Distributed Storage with D2D Communication”. In: [Online]arXiv:1504.06231 (2015)

3/19

Page 12: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

IntroductionSome Relevant Works

I Scaling laws for caching in content delivery networks [GPT12]

I FemtoCaching architecture [Gol+13]

I LivingOnTheEdge: Edge caching and content popularity learningaspects [BBD14]

I Deployment aspects of cache-enabled single-tier networks [Bas+15]

I Coded caching gains in physical layer [ZE15]

I MDS and regenerating codes [BGL15][Ped+15]

[GPT12] Savvas Gitzenis et al. “Asymptotic Laws for Joint Content Replication and Delivery in Wireless Networks”. In: IEEETransactions on Information Theory [Online] arXiv:1201.3095 59.12 (Dec. 2012), pp. 2760–2776

[Gol+13] Negin Golrezaei et al. “Femtocaching and device-to-device collaboration: A new architecture for wireless video distribution”. In:IEEE Communications Magazine 51.4 (2013), pp. 142–149

[BBD14] Ejder Bastug et al. “Living on the Edge: The role of Proactive Caching in 5G Wireless Networks”. In: IEEE CommunicationsMagazine 52.8 (Aug. 2014), pp. 82–89

[Bas+15] Ejder Bastug et al. “Cache-enabled Small Cell Networks: Modeling and Tradeoffs”. In: EURASIP Journal on WirelessCommunications and Networking 1 (Feb. 2015), p. 41

[ZE15] Jingjing Zhang and Petros Elia. “Fundamental Limits of Cache-Aided Wireless BC: Interplay of Coded-Caching and CSITFeedback”. In: [Online] arXiv:1511.03961 (2015)

[BGL15] Valerio Bioglio et al. “Optimizing MDS Codes for Caching at the Edge”. In: [Online] arXiv:1508.05753 (2015)

[Ped+15] Jesper Pedersen et al. “Repair Scheduling in Wireless Distributed Storage with D2D Communication”. In: [Online]arXiv:1504.06231 (2015)

3/19

Page 13: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

IntroductionSome Relevant Works

I Scaling laws for caching in content delivery networks [GPT12]

I FemtoCaching architecture [Gol+13]

I LivingOnTheEdge: Edge caching and content popularity learningaspects [BBD14]

I Deployment aspects of cache-enabled single-tier networks [Bas+15]

I Coded caching gains in physical layer [ZE15]

I MDS and regenerating codes [BGL15][Ped+15]

[GPT12] Savvas Gitzenis et al. “Asymptotic Laws for Joint Content Replication and Delivery in Wireless Networks”. In: IEEETransactions on Information Theory [Online] arXiv:1201.3095 59.12 (Dec. 2012), pp. 2760–2776

[Gol+13] Negin Golrezaei et al. “Femtocaching and device-to-device collaboration: A new architecture for wireless video distribution”. In:IEEE Communications Magazine 51.4 (2013), pp. 142–149

[BBD14] Ejder Bastug et al. “Living on the Edge: The role of Proactive Caching in 5G Wireless Networks”. In: IEEE CommunicationsMagazine 52.8 (Aug. 2014), pp. 82–89

[Bas+15] Ejder Bastug et al. “Cache-enabled Small Cell Networks: Modeling and Tradeoffs”. In: EURASIP Journal on WirelessCommunications and Networking 1 (Feb. 2015), p. 41

[ZE15] Jingjing Zhang and Petros Elia. “Fundamental Limits of Cache-Aided Wireless BC: Interplay of Coded-Caching and CSITFeedback”. In: [Online] arXiv:1511.03961 (2015)

[BGL15] Valerio Bioglio et al. “Optimizing MDS Codes for Caching at the Edge”. In: [Online] arXiv:1508.05753 (2015)

[Ped+15] Jesper Pedersen et al. “Repair Scheduling in Wireless Distributed Storage with D2D Communication”. In: [Online]arXiv:1504.06231 (2015)

3/19

Page 14: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Network ModelScenario

I Our scenario...

4/19

Page 15: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Network ModelScenario

mobile userterminal

I N user terminals (UTs) from the set N = {1, . . . ,N}

4/19

Page 16: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Network ModelScenario

mobile userterminal

I M small base stations (SBSs) from the set M = {1, . . . ,M}

4/19

Page 17: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Network ModelScenario

storage unitCache-EnabledBase Station

mobile userterminal

I Each SBS m has storage capacity of Sm

4/19

Page 18: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Network ModelScenario

storage unitCache-EnabledBase Station

downlink/uplink

mobile userterminal

I A wireless link with total capacity of C ′m Mbyte/s

4/19

Page 19: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Network ModelScenario

base stationcontroller

storage unitCache-EnabledBase Station

downlink/uplink

backhaul link

mobile userterminal

I A wired backhaul link with capacity Cm Mbyte/s for SBS m

I Limited backhaul regime, with Cm < C ′m

4/19

Page 20: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Network ModelScenario

base stationcontroller

storage unitCache-EnabledBase Station

downlink/uplink

backhaul link

mobile userterminal

I A library of F contents, where each content f ∈ FI SBSs proactively cache contents from the library F during peak-off hours

I Each content f has a size of L(f ) Mbyte and bitrate requirement of B(f )Mbyte/s

4/19

Page 21: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Network ModelScenario

base stationcontroller

storage unitCache-EnabledBase Station

downlink/uplink

backhaul link

mobile userterminal

I In ordered case, content requests follow a Zipf-like distributionPF (f ), ∀f ∈ F with shape parameter α

I In unordered case, content popularity matrix Pm(t) ∈ RN×F

4/19

Page 22: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Network ModelPerformance Metrics

A request d is called satisfied if the rate of content delivery is equal or higher than the bitrate ofthe content in the end of service, such as:

L(fd )

τ ′(fd )− τ(fd )≥ B(fd ) (1)

where

I fd describes the requested content

I L(fd ) and B(fd ) are the size and bitrate of the content

I τ(fd ) is the arrival time of the content request and τ ′(fd ) the end time delivery

Users’ average request satisfaction ratio is then defined for the set of all requests, that is:

η(D) =1

D

∑d∈D

1

{L(fd )

τ ′(fd )− τ(fd )≥ B(fd )

}(2)

where 1 {...} is the indicator function. Now, denoting Rd (t) Mbyte/s as the instantaneous rateof backhaul for the request d at time t, with Rd (t) ≤ Cm, ∀m ∈ M, the average backhaul load isthen expressed as:

ρ(D) =1

D

∑d∈D

1

L(fd )

τ′(fd )∑t=τ(fd )

Rd (t) (3)

5/19

Page 23: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Network ModelPerformance Metrics

A request d is called satisfied if the rate of content delivery is equal or higher than the bitrate ofthe content in the end of service, such as:

L(fd )

τ ′(fd )− τ(fd )≥ B(fd ) (1)

where

I fd describes the requested content

I L(fd ) and B(fd ) are the size and bitrate of the content

I τ(fd ) is the arrival time of the content request and τ ′(fd ) the end time delivery

Users’ average request satisfaction ratio is then defined for the set of all requests, that is:

η(D) =1

D

∑d∈D

1

{L(fd )

τ ′(fd )− τ(fd )≥ B(fd )

}(2)

where 1 {...} is the indicator function. Now, denoting Rd (t) Mbyte/s as the instantaneous rateof backhaul for the request d at time t, with Rd (t) ≤ Cm, ∀m ∈ M, the average backhaul load isthen expressed as:

ρ(D) =1

D

∑d∈D

1

L(fd )

τ′(fd )∑t=τ(fd )

Rd (t) (3)

5/19

Page 24: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Network ModelPerformance Metrics

A request d is called satisfied if the rate of content delivery is equal or higher than the bitrate ofthe content in the end of service, such as:

L(fd )

τ ′(fd )− τ(fd )≥ B(fd ) (1)

where

I fd describes the requested content

I L(fd ) and B(fd ) are the size and bitrate of the content

I τ(fd ) is the arrival time of the content request and τ ′(fd ) the end time delivery

Users’ average request satisfaction ratio is then defined for the set of all requests, that is:

η(D) =1

D

∑d∈D

1

{L(fd )

τ ′(fd )− τ(fd )≥ B(fd )

}(2)

where 1 {...} is the indicator function. Now, denoting Rd (t) Mbyte/s as the instantaneous rateof backhaul for the request d at time t, with Rd (t) ≤ Cm, ∀m ∈ M, the average backhaul load isthen expressed as:

ρ(D) =1

D

∑d∈D

1

L(fd )

τ′(fd )∑t=τ(fd )

Rd (t) (3)

5/19

Page 25: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Network ModelBackhaul Offloading Problem

minimizeX(t),Pm(t)

ρ(D) (4)

subject to Lmin ≤ L(fd) ≤ Lmax, ∀d ∈ D, (4a)

Bmin ≤ B(fd) ≤ Bmax, ∀d ∈ D, (4b)

Rd(t) ≤ Cm, ∀t, ∀d ∈ D,∀m ∈M, (4c)

R ′d(t) ≤ C ′m, ∀t, ∀d ∈ D,∀m ∈M, (4d)∑f∈F

L(f )xm,f (t) ≤ Sm, ∀t, ∀m ∈M, (4e)

∑n∈N

∑f∈F

Pmn,f (t) = 1, ∀t, ∀m ∈M, (4f)

xm,f (t) ∈ {0, 1}, ∀t, ∀f ∈ F , ∀m ∈M, (4g)

ηmin ≤ η(D), (4h)

where R ′d(t) Mbyte/s describes the instantaneous rate of wireless link forrequest d and ηmin represents the minimum target satisfaction ratio.

We simplify (and solve) this problem by first estimating P, then cachingcontents greedily (represented by X)

6/19

Page 26: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Network ModelBackhaul Offloading Problem

minimizeX(t),Pm(t)

ρ(D) (4)

subject to Lmin ≤ L(fd) ≤ Lmax, ∀d ∈ D, (4a)

Bmin ≤ B(fd) ≤ Bmax, ∀d ∈ D, (4b)

Rd(t) ≤ Cm, ∀t, ∀d ∈ D,∀m ∈M, (4c)

R ′d(t) ≤ C ′m, ∀t, ∀d ∈ D,∀m ∈M, (4d)∑f∈F

L(f )xm,f (t) ≤ Sm, ∀t, ∀m ∈M, (4e)

∑n∈N

∑f∈F

Pmn,f (t) = 1, ∀t, ∀m ∈M, (4f)

xm,f (t) ∈ {0, 1}, ∀t, ∀f ∈ F , ∀m ∈M, (4g)

ηmin ≤ η(D), (4h)

where R ′d(t) Mbyte/s describes the instantaneous rate of wireless link forrequest d and ηmin represents the minimum target satisfaction ratio.

We simplify (and solve) this problem by first estimating P, then cachingcontents greedily (represented by X)

6/19

Page 27: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Network ModelContent Popularity Learning

If sufficient amount of users’ ratings are available at the SBSs, we canconstruct a k-rank approximate popularity matrix P ≈ NTF, by jointlylearning the factor matrices N ∈ Rk×N and F ∈ Rk×F that minimizes thefollowing cost function:

minimizeP

∑Pij∈P

(nTi fj − Pij

)2

+ µ(||N||2F + ||F||2F

)(5)

where

I The summation is done over the corresponding user/content ratingpairs Pij in the training set P

I The vectors ni and fj here describe the i-th and j-th columns of Nand F matrices respectively

I ||.||2F represents the Frobenius norm

I The parameter µ is used to provide a balance between theregularization and fitting the training data

7/19

Page 28: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Big Data PlatformOverview

base stationcontroller

storage unitCache-EnabledBase Station

downlink/uplink

backhaul link

mobile userterminal

I Mobile Traffic: Approximately over 80 TByte of total data flowing in uplink/downlink dailyin the mobile operator’s core network

I The big data platform runs in the operator’s core network, and collects the data fromseveral base stations

I Collected Data: Approximately 7 hours starting from 12 pm to 7 pm on Saturday 21’st ofMarch 2015

I Big Data Platform: Cloudera’s Distribution Including Apache Hadoop (CDH4) version onfour nodes including one cluster name node

I Each node node with INTEL Xeon CPU E5-2670 running @2.6 GHz, 32 Core CPU, 132GByte RAM, 20 TByte hard disk

8/19

Page 29: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Big Data PlatformOverview

base stationcontroller

centralrouter

core router

storage unitCache-EnabledBase Station

downlink/uplink

backhaul link

mobile userterminal

I Mobile Traffic: Approximately over 80 TByte of total data flowing in uplink/downlink dailyin the mobile operator’s core network

I The big data platform runs in the operator’s core network, and collects the data fromseveral base stations

I Collected Data: Approximately 7 hours starting from 12 pm to 7 pm on Saturday 21’st ofMarch 2015

I Big Data Platform: Cloudera’s Distribution Including Apache Hadoop (CDH4) version onfour nodes including one cluster name node

I Each node node with INTEL Xeon CPU E5-2670 running @2.6 GHz, 32 Core CPU, 132GByte RAM, 20 TByte hard disk

8/19

Page 30: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Big Data PlatformOverview

base stationcontroller

centralrouter

database

cluster computing

core router

storage unitCache-EnabledBase Station

Big DataPlatform

downlink/uplink

backhaul link

mobile userterminal

I Mobile Traffic: Approximately over 80 TByte of total data flowing in uplink/downlink dailyin the mobile operator’s core network

I The big data platform runs in the operator’s core network, and collects the data fromseveral base stations

I Collected Data: Approximately 7 hours starting from 12 pm to 7 pm on Saturday 21’st ofMarch 2015

I Big Data Platform: Cloudera’s Distribution Including Apache Hadoop (CDH4) version onfour nodes including one cluster name node

I Each node node with INTEL Xeon CPU E5-2670 running @2.6 GHz, 32 Core CPU, 132GByte RAM, 20 TByte hard disk

8/19

Page 31: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Big Data PlatformOverview

base stationcontroller

centralrouter

database

cluster computing

core router

storage unitCache-EnabledBase Station

Big DataPlatform

downlink/uplink

backhaul link

mobile userterminal

I Mobile Traffic: Approximately over 80 TByte of total data flowing in uplink/downlink dailyin the mobile operator’s core network

I The big data platform runs in the operator’s core network, and collects the data fromseveral base stations

I Collected Data: Approximately 7 hours starting from 12 pm to 7 pm on Saturday 21’st ofMarch 2015

I Big Data Platform: Cloudera’s Distribution Including Apache Hadoop (CDH4) version onfour nodes including one cluster name node

I Each node node with INTEL Xeon CPU E5-2670 running @2.6 GHz, 32 Core CPU, 132GByte RAM, 20 TByte hard disk

8/19

Page 32: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Big Data PlatformOverview

base stationcontroller

centralrouter

database

cluster computing

core router

storage unitCache-EnabledBase Station

Big DataPlatform

downlink/uplink

backhaul link

mobile userterminal

I Mobile Traffic: Approximately over 80 TByte of total data flowing in uplink/downlink dailyin the mobile operator’s core network

I The big data platform runs in the operator’s core network, and collects the data fromseveral base stations

I Collected Data: Approximately 7 hours starting from 12 pm to 7 pm on Saturday 21’st ofMarch 2015

I Big Data Platform: Cloudera’s Distribution Including Apache Hadoop (CDH4) version onfour nodes including one cluster name node

I Each node node with INTEL Xeon CPU E5-2670 running @2.6 GHz, 32 Core CPU, 132GByte RAM, 20 TByte hard disk

8/19

Page 33: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Big Data PlatformOverview

base stationcontroller

centralrouter

database

cluster computing

core router

storage unitCache-EnabledBase Station

Big DataPlatform

downlink/uplink

backhaul link

mobile userterminal

I Mobile Traffic: Approximately over 80 TByte of total data flowing in uplink/downlink dailyin the mobile operator’s core network

I The big data platform runs in the operator’s core network, and collects the data fromseveral base stations

I Collected Data: Approximately 7 hours starting from 12 pm to 7 pm on Saturday 21’st ofMarch 2015

I Big Data Platform: Cloudera’s Distribution Including Apache Hadoop (CDH4) version onfour nodes including one cluster name node

I Each node node with INTEL Xeon CPU E5-2670 running @2.6 GHz, 32 Core CPU, 132GByte RAM, 20 TByte hard disk

8/19

Page 34: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Big Data PlatformData Extraction Process

databasecluster computing

9/19

Page 35: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Big Data PlatformData Extraction Process

databasecluster computing

tra�cmirroring

high-speed

data �ow

9/19

Page 36: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Big Data PlatformData Extraction Process

databasecluster computing

tra�cmirroring

high-speed

data �ow

Control Packets

Data Packets

...

...

1) Collect Raw Data

9/19

Page 37: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Big Data PlatformData Extraction Process

databasecluster computing

tra�cmirroring

high-speed

data �ow

Location/Session FieldsCELL-ID, SAC, LAC, TEID

Content Request FieldHTTP URI

Request Time FieldFRAME TIME

Control Packets

Data Packets

...

...

1) Collect Raw Data

2) Extract Relevant Fields

9/19

Page 38: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Big Data PlatformData Extraction Process

databasecluster computing

tra�cmirroring

high-speed

data �ow

HDFS & MapReduce

Location/Session FieldsCELL-ID, SAC, LAC, TEID

Content Request FieldHTTP URI

Request Time FieldFRAME TIME

Control Packets

Data Packets

...

...

1) Collect Raw Data

2) Extract Relevant Fields

traces-table-temp

3) MatchFields

HTTP URI FRAME TIME TEID ............... ................... ...................... ................... ...................... ................... .......

9/19

Page 39: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Big Data PlatformData Extraction Process

databasecluster computing

tra�cmirroring

high-speed

data �ow

SizeCalculator

(HTTPClient API)

HDFS & MapReduce

Location/Session FieldsCELL-ID, SAC, LAC, TEID

Content Request FieldHTTP URI

Request Time FieldFRAME TIME

SIZEControl Packets

Data Packets

...

...

1) Collect Raw Data

2) Extract Relevant Fields 4) Calculate

Content Sizes

traces-table-temp

traces-table

3) MatchFields

HTTP URI FRAME TIME TEID ............... ................... ...................... ................... ...................... ................... .......

HTTP URI FRAME TIME TEID SIZE ............... ................... ....... ...................... ................... ....... ...................... ................... ....... .......

9/19

Page 40: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Big Data PlatformData Extraction Process

databasecluster computing

tra�cmirroring

high-speed

data �ow

SizeCalculator

(HTTPClient API)

HDFS & MapReduce

Location/Session FieldsCELL-ID, SAC, LAC, TEID

Content Request FieldHTTP URI

Request Time FieldFRAME TIME

SIZEControl Packets

Data Packets

...

...

1) Collect Raw Data

2) Extract Relevant Fields 4) Calculate

Content Sizes

5) Store Processed Data

traces-table-temp

traces-table

3) MatchFields

HTTP URI FRAME TIME TEID ............... ................... ...................... ................... ...................... ................... .......

HTTP URI FRAME TIME TEID SIZE ............... ................... ....... ...................... ................... ....... ...................... ................... ....... .......

9/19

Page 41: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Big Data PlatformTraffic Characteristics

100 101 102 103 104

100

102

104

106

Rank

Nr.

of

occ

ure

nce

Collected traces

Zipf fit, α = 1.36

Figure 1: Global content popularity distribution.

I The popularity behaviour of contents follows a Zipf law withsteepness parameter α = 1.36

10/19

Page 42: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Big Data PlatformTraffic Characteristics II

100 101 102 103 104

0

5

10

15

Rank

Cu

m.

Siz

e[G

Byt

e]

Figure 2: Cumulative size distribution.

I Total catalog size of 17.7451 GByte

I The cumulative size up to 41-th most-popular contents has 0.1GByte of size whereas a dramatical increase appears afterwards

11/19

Page 43: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Numerical ResultsOverview

Table 1: List of simulation parameters.

Parameter Description Value

T Time slots 6 hours 47 minutes

D Number of requests 422529

F Number of contents 16419

M Number of small cells 16

Lmin Min. size of a content 1 Byte

Lmax Max. size of a content 6.024 GByte

B(f ) Bitrate of content f 4 Mbyte/s∑m Cm Total backhaul link capacity 3.8 Mbyte/s∑

m

∑n C′m Total wireless link capacity 120 Mbyte/s

I Ground Truth: The content popularity matrix P is constructed from all availableinformation in traces-table. The rating density = 6.42%

I Collaborative Filtering: The problem in (5) is attempted by first choosing 10% of ratingsfrom traces-table uniformly at random. Then, these ratings are used in the training stage ofthe algorithm and missing entries/ratings of P are estimated via regularized singular-valuedecomposition (SVD)

12/19

Page 44: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Numerical ResultsOverview

Table 1: List of simulation parameters.

Parameter Description Value

T Time slots 6 hours 47 minutes

D Number of requests 422529

F Number of contents 16419

M Number of small cells 16

Lmin Min. size of a content 1 Byte

Lmax Max. size of a content 6.024 GByte

B(f ) Bitrate of content f 4 Mbyte/s∑m Cm Total backhaul link capacity 3.8 Mbyte/s∑

m

∑n C′m Total wireless link capacity 120 Mbyte/s

I Ground Truth: The content popularity matrix P is constructed from all availableinformation in traces-table. The rating density = 6.42%

I Collaborative Filtering: The problem in (5) is attempted by first choosing 10% of ratingsfrom traces-table uniformly at random. Then, these ratings are used in the training stage ofthe algorithm and missing entries/ratings of P are estimated via regularized singular-valuedecomposition (SVD)

12/19

Page 45: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Numerical ResultsUsers’ Satisfaction

0 20 40 60 80 100

40

60

80

100

Storage Size (%)

Req

ues

tS

ati

sfa

ctio

n(%

)

Ground Truth

Collaborative Filtering

Figure 3: Evolution of satisfaction with respect to the storage size.

I With 40% of storage size, the ground truth achieves 92% of satisfaction whereas thecollaborative filtering (CF) has value of 69%

I There is a performance gap between the ground truth and CF until 87% of storage size,which is due to the estimation errors

13/19

Page 46: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Numerical ResultsBackhaul Offloadings

0 20 40 60 80 100

0

50

100

Storage Size (%)

Ba

ckh

au

lL

oa

d(%

)

Ground Truth

Collaborative Filtering

Figure 4: Evolution of backhaul usage with respect to the storage size.

I With 87% of storage size for caching, both approaches offload 98% of backhaul usage

I Popularity-based storage fails to capture these content size aspects on the backhaul usage

14/19

Page 47: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Numerical ResultsTraining Density

0 20 40 60 80 100

6 · 10−2

8 · 10−2

0.1

0.12

0.14

Training Density (%)

RM

SE

Figure 5: Evolution of root mean square error (RMSE) with respect to thetraining density.

I Increasing training density in this setup improves the estimation15/19

Page 48: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Conclusions

I We have presented a proactive caching approach for 5G wirelessnetworks by exploiting large amount of available data on a big dataplatform and employing machine learning tools

I First study on exploitation of big data for caching in wirelessnetworks

I Performance gains depend on storage size and rating density

Interesting future directions of this experimental work are:

I Detailed characterization of the traffic

I Novel machine learning algorithms

I Deterministic/randomized cache decision algorithms which are notpurely based on content popularity and storing most popularcontents

16/19

Page 49: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Conclusions

I We have presented a proactive caching approach for 5G wirelessnetworks by exploiting large amount of available data on a big dataplatform and employing machine learning tools

I First study on exploitation of big data for caching in wirelessnetworks

I Performance gains depend on storage size and rating density

Interesting future directions of this experimental work are:

I Detailed characterization of the traffic

I Novel machine learning algorithms

I Deterministic/randomized cache decision algorithms which are notpurely based on content popularity and storing most popularcontents

16/19

Page 50: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Thanks for your attentionhttp://www.laneas.com/ejder-bastug

[email protected]

Related PublicationsI E. Zeydan, E. Bastug, M. Bennis, M. Abdel Kader, A. Karatepe, A. Salih Er, and M.

Debbah, ”Big Data Caching for Networking: Moving from Cloud to Edge”, IEEECommunications Magazine, Submitted (2016).

I E. Bastug, M. Bennis, E. Zeydan, M. Abdel Kader, A. Karatepe, A. Salih Er, and M.Debbah, ”Big Data Meets Telcos: A Proactive Caching Perspective”, IEEE/KICS Journal ofCommunications and Networks, Special Issue on Big Data Networking-Challenges andApplications, vol. 17, no. 6, pp. 549–557, December 2015.

I M. Abdel Kader, E. Bastug, M. Bennis, E. Zeydan, A. Karatepe, A. Salih Er, and M.Debbah, ”Leveraging Big Data Analytics for Cache-Enabled Wireless Networks”, IEEEGlobal Communications Conference (GLOBECOM) Workshop, San Diego, CA, USA, 2015.

This research has been supported by the ERC Starting Grant 305123 MORE (Advanced Mathematical Tools for Complex Network

Engineering), the SHARING project under the Finland grant 128010 and TUBITAK TEYDEB 1509 project grant (numbered 9120067)

and the project BESTCOM.

17/19

Page 51: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Bibliography I

J.G. Andrews, S. Buzzi, Wan Choi, S.V. Hanly, A. Lozano,A.C.K. Soong, and J.C. Zhang. “What Will 5G Be?” In:IEEE Journal on Selected Areas in Communications 32.6(June 2014), pp. 1065–1082.

Ejder Bastug, Mehdi Bennis, Marios Kountouris, andMerouane Debbah. “Cache-enabled Small Cell Networks:Modeling and Tradeoffs”. In: EURASIP Journal on WirelessCommunications and Networking 1 (Feb. 2015), p. 41.

Ejder Bastug, Mehdi Bennis, and Merouane Debbah. “Livingon the Edge: The role of Proactive Caching in 5G WirelessNetworks”. In: IEEE Communications Magazine 52.8 (Aug.2014), pp. 82–89.

Valerio Bioglio, Frederic Gabry, and Ingmar Land.“Optimizing MDS Codes for Caching at the Edge”. In:[Online] arXiv:1508.05753 (2015).

18/19

Page 52: Big Data Meets Telcos: A Proactive Caching Perspective · Big Data Meets Telcos: A Proactive Caching Perspective Ejder Ba˘stu g , Mehdi Bennis?, Engin Zeydan , Manhal Abdel Kader

Bibliography II

Negin Golrezaei, Andreas F Molisch, Alexandros G Dimakis,and Giuseppe Caire. “Femtocaching and device-to-devicecollaboration: A new architecture for wireless videodistribution”. In: IEEE Communications Magazine 51.4(2013), pp. 142–149.

Savvas Gitzenis, Georgios Paschos, and Leandros Tassiulas.“Asymptotic Laws for Joint Content Replication and Deliveryin Wireless Networks”. In: IEEE Transactions on InformationTheory [Online] arXiv:1201.3095 59.12 (Dec. 2012),pp. 2760–2776.

Jesper Pedersen, Alexandre Grael i Amat, Iryna Andriyavona,and Fredrik BrannstrPom. “Repair Scheduling in WirelessDistributed Storage with D2D Communication”. In: [Online]arXiv:1504.06231 (2015).

Jingjing Zhang and Petros Elia. “Fundamental Limits ofCache-Aided Wireless BC: Interplay of Coded-Caching andCSIT Feedback”. In: [Online] arXiv:1511.03961 (2015).

19/19