Micro Cast Cooperative Video Streaming on Smartphones Lorenzo

Micro. Cast: Cooperative Video Streaming on Smartphones Lorenzo Keller, Anh Le, Blerim Cic, Hulya Seferoglu LIDS, Christina Fragouli, Athina Markopoulou

Outline • Introduction • Related Work • Micro. Cast Architecture • Micro. Cast Algorithm • Implementation in Details • Evaluation • Discussion

Introduction • Motivation – – • Video traffic increasing(3 G and 4 G as well) A group of users want to watch same video within proximity of each other Problem – Each phone’s individual cellular connection may not be sufficient for providing high video quality

Micro. Cast System Intro. • Micro. Cast • Micro. Download – – • Scheduler Two interface for data delivery Micro. NC-P 2 – – Combination of Wi. Fi overhearing and network coding – • all-to-all dissemination scheme for local sharing content Significantly outperforms state-of-the-art P 2 P schemes Micro. Broadcast

Micro. Cast System Scenario

Outline • Introduction • Related Work • Micro. Cast Architecture • Micro. Cast Algorithm • Implementation in Details • Evaluation • Discussion

Related Work • Cooperative Mobile/Wireless Systems • Wi. Fi direct in Android(Ice Cream Sandwich) • Network Coding – Cooperative/Wireless Systems – Implementation – Peer-to-Peer system – Optimization

Outline • Introduction • Related Work • Micro. Cast Architecture • Micro. Cast Algorithm • Implementation in Details • Evaluation • Discussion

Micro. Cast Architecture • Setup – • 3 G to server and Wi. Fi to local users Assumptions – Small number of users(6 -7) – Users know and trust each other – All user are within proximity of each other – All local links have similar rates on average

Micro. Cast Architecture • Micro. Download – • Micro. NC-P 2 – • Distributing segments using local wireless network Micro. Broadcast – • Only runs on one of the phones that initiate download Pseudo-broadcast over Wi. Fi Requester

Architecture

GUI

Outline • Introduction • Related Work • Micro. Cast Architecture • Micro. Cast Algorithm • Implementation in Details • Evaluation • Discussion

Micro. Cast Algorithm

Micro. Cast Algorithm

Micro. Cast Algorithm

Reception Rate Where Rl is Receive rate in local, Rc is Receive rate in cellular, and N is the number of phones

Outline • Introduction • Related Work • Micro. Cast Architecture • Micro. Cast Algorithm • Implementation in Details • Evaluation • Discussion

Architecture Details • Devices • Requester – • Three types of sources: HTTP, file, and content Storage – – • Java API, Android Media. Player(for playback) GUI

Multiple Network Interface • Downlink(3 G or Wi. Fi) • Local cooperation(Wi. Fi or Bluetooth) • Not suggest Wi. Fi+Bluetooth • Challenges from android connectivity manager – Shut down when both interface are activated

Network Coding • Use generation-based network coding • CPU Limitations – NC is CPU intensive operation – Need an efficient way to encode/decode – Java vs native code (C)

Network Coding • • Simple encoding One segment is devided into m packets each of size n. Let B matrix represents all the m packets. • Size of B is mxn. • Let's take identity matrix I of size mxm • Augment I with B ie [B|I] • Randomly generate coefficient matrix R of size mxm • Encoding R*[B|I]

Network Coding • The encoded message is sent to the network. • Let Inv. R = Inverse of R • Decoding: Inv. R x Encoded Matrix = Inv. R x (R x [B|I]) = (Inv. R x R) x [B|I] • =I x [B|I] = [B|I]

Implementing High-Rate Wi. Fi Broadcast • Disadvantages in 802. 11 broadcast mode – Lacks a back-off mechanism – Transmission rate is limited(base rate, 1 Mbps) – Not always adapt on Android phone(wireless driver and firmware limitation)

High-Rate Wi. Fi Broadcast cont. • Possible sol : Pseudo-broadcast(overhearing) • Challenges: – – Android do not support pseudo-broadcast – • Phone do not support the promiscuous mode Overhearing is not available in ad-hoc mode Implement “pseudo-ad-hoc mode”(Fig 4. (b))

Outline • Introduction • Related Work • Micro. Cast Architecture • Micro. Cast Algorithm • Implementation in Details • Evaluation • Discussion

Evaluation • Download time • Video rate • Testbed : – 4 Samsung Captivate 3 Nexus S – 1 G Cortex-A 8 CPU – 512 MB RAM – 6 use Android 2. 3, 1 use Android 4. 0

Micro. Download

Micro. NC-P 2 • Compare to – – • Bit. Torrent-based distributor (Bit. Torrent-Pull) R 2 -based distributor (R 2 -Push) Consider – Clique and star topologies – using UDP packets locally

Bit. Torrent-Pull&R 2 -Rush • Bit. Torrent-Pull – Three main types of messages • • Request • • Bitfield and have Piece R 2 -Push – Two main types of messages • Data

Bit. Torrent-Pull&R 2 -Rush

Evaluation of Micro. NC-P 2

Evaluation of Micro. Cast • 7 phones – 4 has 3 G • Locally UDP • Support up to 20 Mbps UDP traffic • Star topology • Use pseudo-adhoc • 9. 93 MB file

Evaluation of Micro. Cast

Evaluation of Micro. Cast

Evaluation of Micro. Cast

Energy consumption

Evaluation of NC Implementations

Outline • Introduction • Related Work • Micro. Cast Architecture • Micro. Cast Algorithm • Implementation in Details • Evaluation • Discussion

Discussion • Limitations and Extensions – Highly modular – Bluetooth vs Wi. Fi for local sharing • Pros and cons – – Support single and mulit-hop – – Need not to root the phone But not support broadcast Wi. Fi vs cellular for download from server

Outline • Introduction • Related Work • Micro. Cast Architecture • Micro. Cast Algorithm • Implementation in Details • Evaluation • Discussion

Conclusion • • Micro. Cast enables users to watch the same video from the internet It cooperatively uses the resources on all smartphone of the group Three mainly components Results demonstrate significant performance without battery penalty

Outline • Introduction • Related Work • Micro. Cast Architecture • Micro. Cast Algorithm • Implementation in Details • Evaluation • Discussion

Comment • Simple but good idea • Can try WIFI direct • Lack of some evaluation pictures

Q&A • Thanks for listening