Download presentation
Presentation is loading. Please wait.
Published byGerald Maxwell Modified over 8 years ago
1
Data-Centric Systems Lab. A Virtual Cloud Computing Provider for Mobile Devices Gonzalo Huerta-Canepa presenter 김영진
2
Data-Centric Systems Lab. Motivation too heavy application !!
3
Data-Centric Systems Lab. Conventional Cloud Computing (1) offloading return
4
Data-Centric Systems Lab. Conventional Cloud Computing (2) ●Connecting to infrastructure-based cloud is not always guaranteed o expensive price to use 3G or LTE networks ●But, still want to take advantages from cloud o offloading o increasing the level of parallelism
5
Data-Centric Systems Lab. Ad hoc Mobile Cloud Computing (1) ●Detect nearby nodes in stable mode ●Peer to peer connection ●No incentives for users sharing resources o by finding user pursuing the same task o by splitting the elements of the task among them
6
Data-Centric Systems Lab. Ad hoc Mobile Cloud Computing (2) ad hoc mobile cloud we are using the same app !
7
Data-Centric Systems Lab. Design Considerations ●Resource monitoring and management ●Seamless integration with the existing cloud APIs ●A partition and offloading scheme suitable for mobile de vices ●Activity detection to find users of the same or similar go als ●Spontaneous interaction network support ●A memory cache scheme to save intermediate results ●Lightweight and resource friendly architecture
8
Data-Centric Systems Lab. Architecture ●Application Manager ●Resource Manager ●Context Manager ●Offloading Manager
9
Data-Centric Systems Lab. Application Manager ●in chage of launching and intercepting an applic ation at loading time ●modifying an application to add features require d for offloading o modifying the reference performed when an application is executed the first time
10
Data-Centric Systems Lab. Resource Manager ●in charge of application profiling and resource m onitoring on a local device ●a profile is defined in terms of o # of remote devides needed to create a v-cloud o sensibility to privacy and amount of resources neede d for the migration to happen
11
Data-Centric Systems Lab. Context Manager ●context widget o communicate with the sources of information ●context manager o handles the information and extract new contexts ●social manager o store the knowledge regarding relationship between users ●location o for the mobility traces ●# of nearby devices given by P2P component o for the enabling of a cloud from application manager
12
Data-Centric Systems Lab. Offloading Manager ●in charge of sending and managing jobs from th e node to other remote devices o waits for the results to be delivered back to the appli cation ●detecting failures in the execution ●creating protected space
13
Data-Centric Systems Lab. Architecture
14
Data-Centric Systems Lab. Implementation (1) ●Cloud Computing Provider Client o Hadoop MapReduce ●Ad hoc Mobile Cloud Framework o Hadoop APIs o direct download (classes and interfaces) o RPC (map/reduce)
15
Data-Centric Systems Lab. Implementation (2) ●Modifying applications in order to o intercept and replace reference o Javassist enables Java programs to define a new class at runtime and to modify a c lass file when the JVM loads it ●Communication between devices o XMPP - Yaja! (Yet Another Jabber API) Serverless messaging ●discovery and msging among devices w/o the need of infra Jabber RPC ●XML-PRC
16
Data-Centric Systems Lab. Evaluation - setting Input data sizeless than 100kb Cloud ProviderComposed of 4 servers Virtual Machine MobileJamVM on jailbroken iPod touch Cloud computing providerOpenJDK VM 6 Version of hadoop0.18 Communication between mobile devicesad hoc WiFi accessing the servers802.11b/g compatible AP ApplicationKorean OCR
17
Data-Centric Systems Lab. Evaluation - Result Reason of decrease in performance ●RPC + waiting time ○but saving in processing time ■energy saving ●Small input data size ○disadvantages in hadoop ■new JVM created in each map processing ■memory problem caused by data table ■scanning time at start up
18
Data-Centric Systems Lab. Conclusion and Future works ●Preliminary design for a framework to crate Ad hoc cloud computing providers ●Mobility trace to create stable community ●Context awareness for fault tolerance
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.