Exploiting an Elastic 2-Tiered Cloud Architecture for Rich Mobile Applications
Exploiting an Elastic 2-Tiered Cloud Architecture for Rich Mobile Applications M. Reza Rahimi, Nalini Venkatasubramanian Motivation and Problem Statement Space-Time Workflow• Smart Phones have limited resources such as • To model mobile applications on 2-Tier Cloud battery, memory and computation. Tier 1: Public Cloud architecture, workflow concept from SOA will be used.• One of the main bottlenecks in ensuring mobile (+) Scalable and Elastic • It consists of number of Logical and Precise steps known (-) Price, Delay QoS is the level of wireless connectivity offered by as a Function. last hop access networks such as 3G and Wi-Fi. • Functions could be composed together in different• 3G exhibits: patterns. k Wide area ubiquitous connectivity. Tier 2: Local Cloud But long delay and slow data transfer. (+) Low Delay, Low F1 F2 F3 F1 RTT:• Wi-Fi exhibits: Power, Almost Free 3G Access ~290ms SEQ LOOP low latencies/delays (-) Not Scalable and Point Elastic scalability issues; as the number of users Wi-Fi Access 1 F3 P1 F3 increases the latency and packet losses Point F1 F4 F1 F4 increase causing a decrease in application F2 F2 RTT: 1 P2 performance. ~80ms• 2-Tier Cloud Architecture could increase the QoS AND: CONCURRENT FUNCTIONS XOR: CONDITIONAL FUNCTIONS of Mobile Applications. • We extend this concept to Space-Time Workflow to model mobile applications. • In this optimization problem our goal is to maximize the minimum saving of power, price and delay of the mobile applications. Cloud Resource Allocation for Mobile Applications (CRAM)• Optimal resource allocation for mobile applications is NP-Hard, so heuristic is System Evaluation and Future Directions needed.• CRAM uses the combination of two main • 2 different classes of mobile applications Mobile Applications Ecosystem best practices in heuristic: will be considered: Simulated Annealing (Catching Global Optima) Intensive Computing and Storage, Single User Multiple Users Single User Multiple Users Greedy Approach(Catching Local Streaming Application, Simulated Optima) Annealing • Application profiling and simulation will be Non-Streaming• It uses the following observation for used to evaluate CRAM. Applications pervasive environment that: near user • When the number of users is high CRAM resources usually have better QoS. could achieve 85% of optimal solution.• Need Efficient way to retrieve information • In future CRAM will be extended to use of services on cloud in specific region. efficiently user trajectory. Example Query: “Retrieve all MPEG to Single User AVI decoder services in distance R of mobile user “• R-Tree is an efficient way to answer these Streaming queries. Applications Multiple Users R2 R1 R S2 S8 S1 R1 R2 S6 S4 R3 R3 R4 R5 R6 R5 R4 S3 R6 S1 S1 S2 S5 Low Computation High Computation S5 S8 S3 S4 S7 or Storage or Storage S7 S9 S11 S11 S6 S9 S10 S10 School of Information & Computer Sciences, University of California, Irvine.