Transcript First Year Results

教育部「大學學術追求卓越發展計畫」
下一世代資訊通訊網路尖端技術及應用
分項計畫四
無障礙網路技術的研究與發展
計畫主持人
清大電機系許雅三教授
清大資工系金仲達教授
What Is Effortless Networking?
Easy and effortless access to
network-oriented services
1
The Big Picture
Service Generation
Shopping
Mail Distance
Learning
VOD
Banking
E-commerce
Chat
Service
Delivery
Service Consumption
smart
dust
2
Key Technologies
•Cluster, routing
•Middleware and software
infrastructure
•Edge service
Shopping VOD
Distance
Learning E-commerce
Chat
•Embedded OS
•Service discovery
•Software infrastructure
3
Research Issues
Smart devices for effortless networking
 Heterogeneous, pervasive, huge scale
 Personalization and group interaction
Edge services
 Infrastructure service:

Request/service routing, latency reduction
 Intermediary service:

Adapting and transformation
Service composition and inter-operation
•Pervasive computing
•Quality of service
•Distributed & peer-to-peer computing 4
Research Agenda
Smart devices with effortless-networking
capabilities
Software infrastructure for effortless
networking
Edge services: infrastructure/intermediary
 Transparent working environment for thin-
client/server computing model
 Active edge server
Computation platform: cluster servers
5
Smart Devices for Effortless
Networks
鍾葉青教授
逢甲大學資工系
Research Overview
Research goals:
 To research and develop smart devices
capable of supporting effortless networking
Research items:
 Embedded OS modules for fast development
of embedded OS for smart devices
 Graphics user interface to support wide-range
of smart devices
 Client applications for smart devices
7
First Year Results
Embedded OS modules:
 Modify and modularize Linux kernel
 Port the Linux kernel to WebPhone :
Simulation Test (完成)
 Physical Porting (Should be done at the end of
April 2001)

Graphics user interface:
 Proprietary GUI: based on microwindow and
study how to integrate with Linux kernel

A proprietary GUI for SA1100 simulator
 Cross-platform general GUI: can display
information on different IAs

A Cross-platform GUI for SA1100 Simulator
8
First Year Results (cont.)
Client application: WebPhone
 Embedded OS modules:
Prototype of an embedded OS for SA-1100
mother board (Should be done at the end of
April 2001)
 Develop kernel program library for SA-1100
mother board

 Graphics user interface:
Prototype of proprietary GUI (Should be done at
the end of July 2001)
 Prototype of cross-platform GUI (Should be
done at the end of July 2001)

 Client application:

Prototype of a web browser for WebPhone
(Should be done at the end of July 2001)
9
Plans for the Second Year
Embedded OS modules:
 Port the Linux kernel to PDAs
 Develop SDK for the PDA
Graphics user interface:
 Proprietary GUI for PDA
 Cross-platform general GUI for PDA
Client application: PDA
 Embedded OS kernel
 Prototype of a WAP web browser for PDAs
 GPS capability
10
Research on Java Software
Infrastructures for Pervasive
Computing Environments
李政崑教授
清華大學資訊工程系
Research Overview
Research goals:
 To research on Java software infrastructures for
supporting effortless networks (pervasive
computing environments)
 Bytecode optimizer and performance tuning
and acceleration for Java-based programs
Research items:
 Java Infrastructures at client sites


KVM-like infrastructures
JINI over bluetooth environments
 Java Infrastructure Research at server sites



EJB on clustering environments
Runtime component builders
PSE for component software
 Bytecode optimizer, performance accelerator
12
Java Research Issues at Client Sites
KVM-like infrastructures
 A VM for small devices
 based on open-source solutions
JINI
 Service discovery
 Over mobile environments (bluetooth)
 Fast messages
 Resource scheduling
13
Java Research Issues at Server Site
EJB on clustering environments
 Based on open source solutions
 Fast messages
 Resource scheduling for session and entity
objects
PSE for component software
Runtime component builders
Caching objects and active object
supports for WAN-level supports of
component services
14
Java Performance Acceleration
Bytecode optimizer
Component-level optimizers
Compiler Infrastructures
15
First Year Results
“Runtime Compositions and
Optimizations of Parallel JavaBean
Programs on Clustered Environments” ,
PDPTA ‘2000, Cheng-Wei Chen, Chung-Kai
Chen, Jenq-Kuen Lee, June 26-29, 2000.
 Topics on JavaBean and component object
server: runtime composition and optimizations
of parallel javabean programs
 Useful for component object sever and server
technologies for pervasive computing
 Serialization, externalization, and component
specifications are employed
 Experimental results done on IBM SP-2
16
First Year Results (cont.)
“A Bytecode to Bytecode Optimizer for
Performances”, LCPC ‘2000, Jian-Zhi Wu
and Jenq-Kuen Lee (Aug 9-12)
 A bytecode-to-bycode optimization scheme
for Java stack caches of ITRI CCL Java proc.
 Potentially useful for client sites with java cards
and java processors in a pervasive computing
environment
 Extension works are on the way and are filing
for patents via ITRI CCL
 The framework can be useful for other types of
java optimizations which can result in
performance improvements for both server
17
and client sites with Java programs
First Year Results (cont.)
Integration of JINI and bluetooth:
(Ericsson Bluetooth Development Kit)
18
JINI and Bluetooth
Jini needs devices
Devices go wireless
 Bluetooth
needs a
more powerful
distributed
environment
Internet
LAN
Internet
LAN
19
Initial Steps for JINI over Bluetooth
Use Jini to control a device with no JVM
– JNI: Java Native method Invocation
JNI & connection
cable/ Wireless)
Jini proxy server
20
First Year Activities
Help to collaborate and co-organize the
Seminar on “Mobile Computing
environment: Device, protocol and
application software”, Nov. 29, 2000.
Organize ASPEISA ’00 (Advanced Software
for Pervasive Environments and Information
and Server Appliances workshop), Dec.
29, 2000.
21
Plans for the Second Year
Java wrapper for protocol stacks of
bluetooth (HCI, L2CAP, RFCOM, SDP)
Java RMI over bluetooth
Java RMI over VIA (component server
technologies)
Runtime compositions of Javabean
components
Research on specification issues on
service discovery.
22
MAS TC/S Architecture for
Pervasive Computing
黃冠寰教授
國立暨南國際大學資管系
Research Overview
Research goals:
 To construct a transparent working
environment for roaming thin-client across
wide area network (WAN)
 Multiple-Application-Server Thin-Client/Server
Computing (MAS TC/S) Architecture
24
First Year Results
We have finished the infrastructure to
support a transparent working
environment for thin-client/server
computing model
 A thin-client is able to roam around WAN
 Multiple-Application-Server Thin-Client/Server
Computing (MAS TC/S) Architecture
A prefetch mechanism for transparent working
environment
 Data classification scheme to support
intellectual data prefetch
 Workflow model to support interactive data
prefetch
 Service discovery and redirection protocol

25
First Year Results (cont.)
We are implementing the MAS TC/S
architecture for demonstration:
 The platform of application server is



Linux + X protocol
Star Office
A GPL X protocol Applet
 Linux OS is modified for a transparent data
prefetch
 Service discovery server
 Working data prefetch daemon
 Task prefetch daemon
26
First Year Results (cont.)
Preliminary result of multiple-application-
server topology is published in Workshop
on Advanced Software for Pervasive
Environments and Information and Server
Appliances (ASPEISA '00)
A full paper is in preparation for submitting
to conference and journal
27
Plans for the Second Year
Apply data mining technique to predict
the working data for prefetch
Design the coherence and cache
protocol for data (files) among multiple
application servers
28
Proposed Research: Next 3 Years
Extend the MAS TC/S architecture to
wireless environment
Establish the Pervasive Network Service
(PNS) environment
 Users can get services by wired or wireless
hardware thin-client devices
 Services are not limited to information but also
execution of application code
=> no limitation to application software:
(1) Legacy Win32 code, newly developed Java
applications, KVM packages for PDA
(2) A small program with 2KB or a large software
with 200MB memory requirement are both able
to serve a small handheld thin-client device
29
Active Edge Servers
金仲達教授 清華大學資訊系
廖維國教授 交通大學電信系
Research Overview
Research goal:
 To research and develop platforms for efficient
edge services
Research topics:
 Infrastructure service for latency reduction

Caching, prefetching, service discovery, ...
 Intermediary service for manipulating client-
server information flow

Transforming, adapting, filtering, …
 Rapid new service development, composition
 Cluster support
 Issues: QoS guarantee, wireless accesses,
peer-to-peer/distributed processing, interproxy operations
31
First Year Results
clients
4. Choosing Path
Front End
1. Request
Handle i
3. Request
2. look up
Client i
Handle i
table
Document
Editor
5. Request
Request
Editor
Cache
6. Find Hit
7. Request
Generator
8.a Get data
8.b Get data
9. tailoring
Pre-fetch
Edge Server Platform
cluster
Monitor
servers
Plan for the Second Year
Prototype of edge server platform
 Resource management and differentiated
services for QoS guarantee
 Supports for peer-to-peer networking, service
discovery
 Supports for wireless accesses and mobile
clients
Applications
 Image transcoding
 Web mail
 Wireless gateway
33
Cluster System Supports
許雅三教授 清華大學電機系
金仲達教授 清華大學資訊系
Research Overview
Research goals:
 To study high performance and scalable
servers for internet services
Research topics:
 Cluster system design for internet services




Network and system scalability
Content-aware service dispatch
Load balancing
Service differentiation and QoS guarantee
 High speed network interface (VIA) and its
supports for internet services
35
First Year Results
Content-based music retrieval on clusters:
 Port a CBMR system to a cluster
 Initial performance evaluation
request
result
Client
Web server
(switch)
Front-end
Back-end
36
Plans for the Second Year
Design a cluster-based and QOS-
supported system for general web services
 Improve dispatch policy to achieve load-
balancing on backends
 Resource management and task scheduling
for service differentiation, performance
isolation
 Use content-based music retrieval as the
target application
37
A New Kind of Internet
-- Randy Katz, UC Berkeley
Service- rather than applications-centric
Programmable intelligence inside the net,
but intelligence shifted to network edges
 Infrastructure proxies intermixed with switches



High local b/w, wide-area b/w efficiency
Fast response time (and predictable)
Integrate localized content
 Rapid new service development
Pervasive support for services
 Personalization, nomadic
 Automatic replication, redirection
 Data transformation, adapting, and filtering
38