Personal Content Caching for Mobiles

Transcription

1 Personal Content Caching for Mobiles Roy Yates Shweta Jain () Ryoichi Shinkuma (Kyoto University)

2 CNF Network Architecture Opportunistic Transmission & Storage Routing Store Cache popular content Movie1. mpg R 1 R 2 R 3 Hop by Hop Transport Media Server Mobile Client 2

3 Caching/Storage Router In-network storage Temporary storage of content due to disconnection, poor link quality, congestion etc Buffering for hop-by-hop transit Personal content Intermittently connected mobile users

4 Disconnected Operation: Post Office Movie1. mpg M1 D C B File Cached Disconnection: hold the file A Inform PO Netflix.com Post Office Mobile Content Location M1 Movie1 C Cache And Forward (CNF) Network Architecture and Design

5 Mov ie1. mp g M1 Cache Space is a Network Resource C D B Post Office In an ideal network: Unlimited cache storage Files cached everywhere On reconnection: file delivered from best cache Max delivery probability Min network load A Netflix.com In practice: Finite capacity caches Personal content is hard to track Cache space easily wasted Cache And Forward (CNF) Network Architecture and Design

6 System Model Mo vie 1.m pg D C B Mobile users connect via access networks request personal content stored in the network. Post Office A Netflix Before content delivery mobile moves and disconnects and later reconnects at a new location What happens to content in transit? Which routers should cache the file? How should old content be discarded? Cache And Forward (CNF) Network Architecture and Design

7 Personal Caching Issues Caches need to be selective File still in cache when mobile reconnects? ELT = Expected LifeTime of a file in a cache M = average mobile away time Cache useless if ELT << M Cache rules need to be local Global network state is too complex Where is this file already cached? Where are all the other files cached? What are the mobility characteristics of this user?

8 Personal Caching Approach Least Recently Used (LRU) cache deletion Identify cache storage rules so LRU works

9 Cache Storage Policies? Cache everything at every intermediate router Last node caching at last router when mobile disconnects Random caching Cache Pricing

10 Random Caching Node i caches a file with probability c i Cache Probability Optimization Node i cache capacity = N i files File arrival rate λ i Little s Law: ELT i = N i /(c i λ i ) Maximize c i such that ELT M M = Mobile away time

11 Cache Pricing User j specifies acceptable price q j for caching Acceptable prices are random over users Router i sets a caching cost p i User j file cached at router i if p i q j Router i sets min p i such that ELT M.

12 Cache Pricing Enables User Selectivity Short stay-time users: often fail to retrieve their files will get bigger benefit from nearby caches will have high willingness to pay for cache 12

13 Simulation Setup T T T T 8 stub networks ring transit network Stub link: 100Mbps, transit link: 1Gbps File size: 1GB, Storage size: GB Stay time: uniform ( sec) Away time: exponential (mean=6000 sec) Probability of movement to other stubs: 0.8

14 Cache Pricing: Short-Stay users pay more large No. of retrieval attempts small Stay time 1000 Acceptable price Stay time 14

15 Retrieval attempts No. of retrieval attempts NoCache Ideal Last Every CP Price Storage size 16

16 Cache retrieval ratio 1.0 Cache retrieval ratio NoCache Ideal Last Every CP Price Storage size 18

17 Discussion Your mileage may vary Results depend on network topology, user mobility, Users Cache pricing policies needs more study User/Router Cache price interaction/equilibrium Cache pricing + last-router caching? Local neighborhood cache coordination? File-embedded caching instructions 19

18 SIMULATION RESULTS Dots represent adjusted pricing, lines represented uniform pricing Cache Hit Ratio