Slide 1

Architectures and Systems for Mobile-Cloud Computing: A Workload-Driven Perspective Prashant Nair Adviser: Moin Qureshi ECE Georgia Tech Xin Zhang Adviser: Mayur Naik CS Georgia Tech S2014-6613 3/26/2014 Silicon Valley

Slide 2

Motivation Mobile devices have become the primary computing device years performance 3/26/2014 Silicon Valley 2G 3G 4G 2

Slide 3

Hello Siri, What is Mobile Cloud Computing? Call John. Call John Dialing 123-456-7890 33/26/2014 Silicon Valley

Slide 4

Mobile-Cloud: Enabling New Applications Idea: Offload Computation to Cloud 43/26/2014 Silicon Valley

Slide 5

Key Challenges 0101010 1010101 1111010 1110001 1. Interleaved I/O and computation 2. Network latency 0101010101 1010100101 1100000101 1010101101 3. Diverse and dynamic execution environment 53/26/2014 Silicon Valley

Slide 6

Our Solution: Flexible Offloading Schemes Bi-directional offloading 1101000 1000101 11000 0101010 1010101 01010 63/26/2014 Silicon Valley Challenge 1: Interleaved I/O and computation

Slide 7

Our Solution: Lower Bandwidth via Persistence Persistent cloud 0101010 1010101 1101010 1000101 01010 0001110 0010101 Challenge 2: Network latency 73/26/2014 Silicon Valley

Slide 8

Our Solution: Analytical Models for Tradeoffs Software features Network features Hardware features Analytical model Runtime decision! Challenge 3: Diverse and dynamic execution environment 83/26/2014 Silicon Valley

Slide 9

Offloading System Analytical model How to offload Offloading schemes What to offload 93/26/2014 Silicon Valley

Slide 10

Roadmap Mobile workload tracing Trace mobile workloads of top 150 Google Play apps Workload characterization Identify features common and unique to mobile workloads Analytical models of performance and energy usage Produce tolerable error bounds compared to hardware measurement Mobile-cloud computing system Show speedup and energy savings (Infrastructure implemented) (~3 months) (~6 months) 103/26/2014 Silicon Valley

Slide 11

Result of Tracing Chess For 10 Rounds UI AI 24 threads 3 million function calls 17 million memory reads 13 million memory writes 113/26/2014 Silicon Valley

Slide 12

Summary Mobile devices have become the new PC Big performance gap between mobile and desktop Our Proposal: Enable desktop-class performance for mobile apps by: Offloading computation to the cloud Using robust analytical modeling Enable new applications and usage models 123/26/2014 Silicon Valley