Presentation is loading. Please wait.

Presentation is loading. Please wait.

Computation Offloading

Similar presentations


Presentation on theme: "Computation Offloading"— Presentation transcript:

1 Computation Offloading
11/13

2 Proposal - Offloading Identify essential information for making offloading decisions Essential information retrieval What to identify How to identify Implement/Improve framework How to build our prototype The influences of property change Computation-intensive => I/O-intensive Decision policy Client OS/Server side/Hybrid offloading decision making. Smart offloading policies. (Client OS rescheduling) Resource integration Virtual storage Support file-related computation-intensive applications. Ex: Face matching /proc

3 Essential Information Retrieval
What to identify? According to most related works, they care about: Power consumption Performance Execution time Response time

4 Related Works Cyber foraging
a pervasive computing technique where resource poor, mobile devices offload some of their heavy work to stronger surrogate machines in the vicinity. Framework for Power Aware Remote Processing, ‘01 Puppeteer, ‘01 Spectra, ‘02 Chroma, ‘03 Goyal & Carter's system, ‘04 Slingshot, ‘05 Scavenger, ‘09 Pervasive 無處不在的 Surrogate 代孕的 Vicinity 附近

5 Related Works(Cont.) Scavenger Offloading Python applications
Remote data store Client-side scheduling Considers surrogate strength and utilization, input size, network media(speed), and data locality. Can add output size and task complexity into consideration. Local and global profiling. Application developer needs to annotate the offloading function.

6 Related Works(Cont.) CloneCloud
Uses a combination of static analysis and dynamic profiling to optimally and automatically partition an application. Optimization objective is to choose a set of partitions that minimize sum of computation cost and migration cost. Use a standard integer linear programming solver to solve the optimization problem.

7 Related Works(Cont.) Kumar’s work takes into account:
The speed of both smart phone and remote cloud resource. The number of bytes that need to be transferred. The network bandwidth. The energy consumption of the smart phone in idle, computing, and communicating state.

8 Related Works(Cont.) MAUI Offloading .NET applications(C#)
Developer annotates “remoteable” methods. Profiling Device: energy consumption. Program: state transfer requirements, runtime duration, # of CPU cycles. Network: Round-trip time, bandwidth, pack loss. Also formulate into a 0-1 ILP problem. Solver is on the server instead of device. Device: Energy consumption of a method as a function of # CPU cycles

9 Summary Most of the existing works focused more on power than performance. The most important element is the network speed, then the computation ability of server. I think we can start by monitoring following metrics: (device)CPU load changing Network speed 3G/Wifi

10 Applicable Image processing Audio processing Text processing
object recognition, OCR, face detection, barcode analysis Audio processing speech recognition Text processing machine translation Artificial intelligence for games Chess 3D rendering 3D home interior design Security taint analysis and virus scans

11 My Consideration The cyber foraging technique starts from 2001.
Why it does not become popular during the rise of smart phone?


Download ppt "Computation Offloading"

Similar presentations


Ads by Google