Transcript Exploiting Purdue`s Technological Prowess

Purdue-Tivoli Partnership:
Exploiting Purdue’s Technological
Prowess
Walid G. Aref
Associate Professor, CS. Dept.
XML Databases and Data Mining
Purdue-Tivoli Partnership
1
Database Systems Research

The objective of my research is:
1.
Build efficient database engines for new data
types:
1.
2.
3.
2.
spatial/geographical databases
semi-structured/unstructured web databases
multimedia databases
Develop algorithms to answer new database
query types:
1.
2.
data mining algorithms
spatial query processing algorithms
Purdue-Tivoli Partnership
2
Proposed Projects


Building an XML-based Database System
Prototype
Data Mining of Event Traces in a
Distributed System
Purdue-Tivoli Partnership
3
XML &Web Databases


XML as the Internet uniform language for
information interchange
Target:


Seamless integration of databases/non-information
sources
view/query the web as a huge database


Querying web structure and contents
personalize/adapt based on customer’s needs/patterns
Purdue-Tivoli Partnership
4
XML/Web Databases: Phase 1




Design data model for XML document database
Design algebra and query language for XML
data model
Build prototype XML database engine
Design and prototype XML views for


legacy relational database systems
HTML pages
Purdue-Tivoli Partnership
5
XML &Web Databases: Phase 2




Develop indexing techniques in XML databases
Develop query processing and optimization
techniques for XML queries
Maintenance of user profiles for personalized
views and query answers
Some applications:



Web mining: Clustering web users based on access
patterns
Web/site searching and mapping to an XML database
modeling distributed system’s topology and resources
using XML
Purdue-Tivoli Partnership
6
XML &Web Databases: Project Plan

Phase 1 deliverables (duration: 18 months)




Phase 2 deliverables (duration: 18 months)




prototype XML database system
search capability + views to relational db/HTML documents
End-of-Phase-1 project report + possible publications/patent
filings
efficient prototype XML database system with
indexing/query processing and optimization capabilities
web site mapping into an XML database and db/web query
engine with user profiling and web mining capabilities
Final project report + possible publications/patent filings
Estimated dollar value: $150K
Purdue-Tivoli Partnership
7
Data Mining of System Event Traces

Target:






Find common event sequences
detect irregular event patterns
predict future events and attempt to prevent them
Mining common event sequence patterns
Mining user actions and responses
Take into consideration system topology and
structure
Purdue-Tivoli Partnership
8
Data Mining of System Event Traces:
Phase 1




Develop understanding of the Tivoli suite of
Enterprise Management Software
Get sample traces of events, user actions, along
with the corresponding sample system/network
topologies
Analyze the data and design the needed
structures and schemas
Build a prototype data warehouse for event
traces, actions, and topologies
Purdue-Tivoli Partnership
9
Data Mining of System Event Traces:
Phase 2





Apply existing data mining techniques
Develop new algorithms for mining the traces
given the system/network topologies
Developing incremental data mining techniques
of event traces
Analysis of the data mining results
Iterate through the above process
Purdue-Tivoli Partnership
10
Data Mining of System Event Traces:
Project Plan

Phase 1 deliverables (duration: 18 months)



Phase 2 deliverables (duration: 18 months)





prototype data warehouse of distributed system event/action
traces and system topologies
Phase-1 project report + possible publications/patent filings
prototype data warehouse with data mining capabilities
new incremental algorithms for mining traces considering
topologies
sample study, results of mining the traces, and
recommendations
Final project report + possible publications/patent filings
Estimated dollar value: $110K
Purdue-Tivoli Partnership
11
Past Research Projects

Distributed Scalar Storage Servers






Using Network-Attached Storage Devices (NASD)
Design and prototype of a distributed real-time file
system (multimedia server)
Past research with Panasonic/Matsushita
(1 European patent granted, 3 U.S. patent filings)
Several journal and conference publications
Data mining in time-series databases


in collaboration with IBM Almaden, San Jose
two patent filings
Purdue-Tivoli Partnership
12
Ongoing Research Projects

Multimedia database systems (On-going)

indexing techniques





textual/caption annotation
retrieval by content
data mining of multimedia data
ACM SIGMOD 95, IEEE ICDE 95, 3 U.S. patents
Spatial database systems (On-going)

Prototype systems



Sand: Prototype spatial database system (University of
Maryland, College Park, 1990)
Spatial index attachment in Starburst extensible DBMS
(IBM Almaden Research, San Jose, CA, 1992)
Program committee member VLDB 2000 (Intl. Conf.)
13
Purdue-Tivoli Partnership
Purdue-Tivoli Partnership:
Exploiting Purdue’s Technological
Prowess
Elisa Bertino
Professor
Access control mechanisms for XML
document sources
Purdue-Tivoli Partnership
14
Long Term Objectives

The objective of my research is:



Development of tools supporting the
specification of access control policies for
XML documents
Development of access control mechanisms
for XML documents
Development of automatic classification
tools for XML documents
Purdue-Tivoli Partnership
15
Impact




The system we develop will support the access control
administration for heterogeneous sources of XML documents
The system will support a language for a high-level definition
of access control policies
The system we develop will support import/export of XML
documents among different sources
It will allows one enable selective distribution of documents to
large user communities
Purdue-Tivoli Partnership
16
Research Methodology

We plan to develop the access control system
and the administration environment on top of
a DBMS supporting XML
Purdue-Tivoli Partnership
17
Research Plan







Definition of an access control model for XML
documents
Definition of an XML library for the encoding of access
control policies and authorizations
Integration of the Access Control Model with User
Credentials mechanisms
Implementation of the Access Control Model
Secure Dissemination of XML Documents
Extension of the Access Control Model to deal with
Multimedia Data
Development of Access Control Policies for Specific
Applications (such as workflow systems)
18
Purdue-Tivoli Partnership
Past Research



Formal definition of an authorization model for XML documents.
Automatic classification of semi-structured and XML documents
Relevant papers:
E.Bertino, et Al, “An Approach to Classify Semi-Structured
Objects”. Proc. ECOOP 99.
E.Bertino, et Al. “Controlled Access and Dissemination of XML
Documents”. Submitted for publication.
E.Bertino, et Al. “An Approach for the Specification and
Enforcement of Authorization Constraints in Workflow
Management Systems", ACM Trans. On Information and Systems
Security, Vol.2, No.1, pp.65-104, February 1999.
Purdue-Tivoli Partnership
19
Purdue-Tivoli Partnership:
Exploiting Purdue’s Technological
Prowess
William J. McIver, Jr.
Visiting Assistant Professor*
Souk Nets: A Component-based
Database Integration Paradigm
Purdue-Tivoli Partnership
20
Long Term Objectives

The objectives of my research are to:




Develop a component-based paradigm
tailored to database integration.
Design a language to use within this
paradigm.
Develop a set of components for
implementing database integration solutions.
Optimize component container approaches
for database applications.
Purdue-Tivoli Partnership
21
Impact

Allow data integrators to leverage benefits of
component-based software construction:




Allow prefabricated functionality to be reused.
Perform more robust reuse.
Perform modular checking in the face of evolution.
Produce a Component-based Paradigm tailored to
data source integration:

Current component-based approaches (e.g. EJBs) are lacking in
this domain.



Too low-level
Imperative & tedious
Current containers are inefficient for database access.
Purdue-Tivoli Partnership
22
Impact

Enable better construction of database integration
solutions:







Rapid
Robust
Reusable
Analyzable
Fault Tolerant
Evolvable
Enable the construction of more efficient
based database applications.
Purdue-Tivoli Partnership
component-
23
Research Methodology

Conceptual


Identify canonical use cases for this technology.
Factor the domain of data integration solutions.



Design a covering set of components to implement these
solutions.
Explore use of reflection, contracts, design patterns, and metadata approaches.


Federation & Schema Integration, Global Query Language
Approaches, Point Solutions, etc.
Build in reasoning capabilities for composition, changes to
interfaces, fault situations.
Design a high-level, cross-platform language.
Purdue-Tivoli Partnership
24
Research Methodology

Theoretical


Extend the LINDA notion of Tuple Space
(Gelertner & Carriero)
 Accomodate complex objects
 Object Spaces
Employ WoFNets Semantics (Ellis & Keddara 1999) to
interconnect Object Spaces.
 A Variant of Colored Petri nets
 Provides a formal semantics
 Supports dynamic change
 Applicable to handling evolution of requirements
 Possible applicability to the active networks paradigm
Purdue-Tivoli Partnership
25
Research Methodology

Theoretical (continued)

Transitions in Souk nets constitute components.
 Structural and value transformations
 Filters
 Control flow
 Event subscription, notification and handling
 User defined transitions
Purdue-Tivoli Partnership
26
Research Methodology

Experimental



Evaluate modeling capabilities of paradigm and language.
 Employ selected use cases.
Conduct performance evaluations of run-time system.
Iterate on language design and system implementation.
Purdue-Tivoli Partnership
27
Research Plan

Schedule & Milestones



Year 1

Identify a canonical set of use cases for database integration.

Implement baseline prototypes of use cases for analysis.

Complete the first version of the component paradigm.
Year 2

Implement an environment based on the component paradigm.

Begin an iterative evaluation process (through Year 3).
Year 3

Implement a revised environment.
Purdue-Tivoli Partnership
28
Research Plan


Deliverables

Software artifacts from each milestone

Results of each milestone reported in appropriate publications
and conferences

Software demonstrations
Staffing & Budget (estimated)



Principal investigators: 1 to 2 FTE.
Graduate students: 3 to 4.
Budget: $800,000.
Purdue-Tivoli Partnership
29
Past Research

The Sanctuary Project

A mediator-based database integration environment for CORBA and
DCOM-based heterogeneous data sources. Supported by NSF grant
IRI-9632595.

Used to perform data migration from Unidata/VMARK CODASYL
databases to O2 object-oriented databases; construct object-oriented
applications atop CODASYL applications; support the integration of
the object-oriented Catalyst software engineering environment with
ODBC-compliant DBMSs.

John Todd, Roger King, William J. McIver, Jr., Richard Osborne, Christian
Och, Nathan Getrich, Brian Temple. “Building Mediators from Components.”
(To appear) Proceedings of The International Symposium on Distributed
Objects and Applications (DOA’99). Edinburgh, Scotland. September 5 - 6,
1999.
Purdue-Tivoli Partnership
30
Past Research

Souk nets (preliminary work)

Development of initial analytic/conceptual framework for
component-based database integration. (Since April 1999)

William J. McIver, Jr., Karim Keddara, Christian Och, Roger
King, Clarence A. Ellis, John Todd, Nathan Getrich, Richard M.
Osborne, Brian Temple. “An Overview of Souk Nets: a
component-based paradigm for data source integration." (To
appear) The Seventh International Workshop on Database
Programming Languages (DBPL 1999). Kinloch Rannoch,
Scotland. September 1st - 3rd, 1999.
Purdue-Tivoli Partnership
31
Purdue-Tivoli Partnership:
Exploiting Purdue’s Technological
Prowess
W. Kent Fuchs
Head & Professor
Electrical & Computer Engineering
Dependable Distributed & Mobile Computing
Purdue-Tivoli Partnership
32
Long Term Objectives

The objectives of my research are:

Rapid recovery from failures




Hardware and software faults
Clusters of networks
Mobile notebooks and hand-held devices
Accurate and preventive diagnosis of faults
Purdue-Tivoli Partnership
33
Impact

Highly reliable computation and
communication in changing environments
Wired network
Mobile
support
station
Mobile
hosts
Mobile environment
Homogeneous environment
Purdue-Tivoli Partnership
Heterogeneous environment
34
Research Methodology

RENEW –– Recoverable Networks of
WORK
Workstations
* Simple application development
P1 P2
* Good performance
* Transparent fault recovery
P3 P4
W O R K
Application
User
User requiring
dependable
computing
* Rapid prototyping of new FT techniques
* Standard benchmarks
* Representative environments
Purdue-Tivoli Partnership
35
Application
MPI
Message
Passing
Module
Ckp. & Rec. protocol
Job
Fault
Managem. Detection
Process
Ckp.
Server
Ckp.
Operating System
Computing nodes
Operating System
File servers
Ethernet, ATM
Purdue-Tivoli Partnership
36
Exec.Time[sec]
4000
NO Ckp
Coordinated
Comm-Induced
Mesg-Logging
3500
3000
2500
2000
1500
1000
500
0
BT
LU
SP
PCCM2
Seismic1
Seismic4
NOTE: Ckp period 5 min.
Purdue-Tivoli Partnership
37

PREACHES (Portable Recovery and Checkpointing
in Heterogeneous Systems)
X
Program
Local Disk
(sec)
btree
flops20
nsieve
swim
tfftdp
tomcatv
mgrid
1.541
0.007
0.007
0.378
1.677
0.065
0.219
Remote Disk
(sec)
X
Backup
(sec)
Checkpoint Size
(KB)
3.87e-4
3.85e-4
3.83e-4
3.86e-4
3.85e-4
3.83e-4
4.34e-4
58653
52
52
14500
65595
2135
7517
5.668
0.020
0.018
1.401
6.033
0.263
0.857
Purdue-Tivoli Partnership
38
Research Plan

Recoverable Mobile Distributed Systems
 High availability and reliability
 Power and bandwidth conservation
Checkpoint
MSS2
MSS1-MSS2-MSS3
Checkpoint MSS2
MSS3
MSS1
HA
Purdue-Tivoli Partnership
39
Past Research

RENEW (Recoverable Network of Workstations)


N. Neves and W. K. Fuchs, “RENEW: A Tool for Fast and
Efficient Implementation of Checkpoint Protocols,” IEEE FaultTolerant Computing Symposium, pp. 58-67, June 1998.
PREACHES (Portable Recovery and Checkpointing in
Heterogeneous Systems)


K.-F. Ssu and W. K. Fuchs, “PREACHES –– Portable Recovery
and Checkpointing in Heterogeneous Systems,” IEEE FaultTolerant Computing Symposium, pp. 38-47, June 1998.
RAMs (Recoverable Mobile Systems)

B. Yao, K.-F. Ssu, and W. K. Fuchs, “Message Logging in Mobile
Computing, IEEE Fault-Tolerant Computing Symposium,” pp.
294-301, June 1999.
Purdue-Tivoli Partnership
40
Purdue-Tivoli Partnership:
Exploiting Purdue’s Technological
Prowess
Shimon Y. Nof
Professor of Industrial Engineering
Design of Middleware Protocols
for e-Business Interactions
Purdue-Tivoli Partnership
41
Long Term Objectives

The objectives of our research are:

Develop a set of collaborative workflow
protocols for guiding and optimizing the
performance of e-business interactions in
heterogeneous, autonomous and distributed
environments, e.g. network of ERPs, HelpDesks

Develop complex problem-solving
scheme/protocol via interactions among
distributed knowledge-based systems
Purdue-Tivoli Partnership
42
Long Term Objectives (continue)

Design recommendations for knowledgebased protocols customized for the needs of
particular enterprises and markets

Design of an executable specification-tool for
protocol development, which will translate
interactions and flow definition of particular
protocols into executable code, subject to the
needs of the users and the organization
Purdue-Tivoli Partnership
43
Impact

In the emerging global e-business market, effective
service availability, e.g., 24x7; Tivoli’s Service
Desk, is a key to success. The collaborative work
protocol is a task administration protocol which will
provide:

Automation among interactions that include decision activities;
automation of the process to provide service at minimum cost and
maximum quality

From our experimental results, specification and working
environment of protocols will be identified, to design/select the right
protocol for the right situation

In complex decision-making processes, human interactions are
required, e.g Help desk application. The protocol will reduce
decision-making time/cost by extracting the right information
Purdue-Tivoli Partnership
44
Research Methodology

The research will employ a new version of TIE,
Teamwork Integration Evaluation developed previously
with NSF support. TIE’s purposes:

Compute performance measures of protocols





completion time, e.g. transaction processing, negotiation
penalty measures, e.g. # of aborted (time-out) connections
messages queue of each party
relative cost-quality model of the service system
Model the interactions among parties in both
synchronous and asynchronous mode. TIE is based on
the MPI technique, it provides true parallelism
analysis of the interaction behavior.
Purdue-Tivoli Partnership
45
Research Methodology
(Continue)

With MPI, TIE can run on both parallel machine like
Paragon, Origin2000; or network of computers e.g.,
Suns, PC windows-NT

To compare protocols’ performance, experiments will
be conducted under different environments with
variable # of participants and service demand

Use traditional e-business protocol “Contract Net” as
base-line protocol

Use IBM’s Situation Manager to coordinate
conditional, triggered actions
Purdue-Tivoli Partnership
46
Research Plan for Three Years

Develop specifications of protocols for e-business
requirements, based on previous research (4 Months)

Modify TIE for protocol evaluation, conduct protocol
performance experiments, analysis (8 Months)

Develop TIE description language for general modeling
purpose; implement for target application (12 Months)

Develop TIE conversion program for executable protocol
code; apply for target application (12 Months)

DELIVERABLES:

BUDGET:
Protocol models, TIE, Language, Converter
Advisor + 2 students @ $50,000/ year
Purdue-Tivoli Partnership
47
Past Research

DPIEM -- Distributed Parallel Integration Evaluation Model
Organizing/reorganizing resources among distributed
networked organizations, based on parallelism theory of
computing & communication
(Ceroni and Nof, 1999, Research Memo No. 99-04, School of IE,
Purdue University)

ABMS -- Agent-based Manufacturing System
General model of cooperation & collaboration among
autonomous agents, resources and tasks. Shows the need
to use workflow protocol to coordinate agents’ tasks
(Huang and Nof, 1998, Int’l Journal of Production Research)
Purdue-Tivoli Partnership
48
Past Research (continue)

DAF-Net -- Data Activity Flow Net ; and AIMIS -Agent-based Integration Model of Information Systems
A collaboration scheme and coordinated execution for
distributed, heterogeneous CIM data activities
(Kim and Nof, 1998, Int’l Journal of Industrial Engineering)

Active database coordination of multiple CIM databases
Monitors events/situations of interest and, when given
conditions are met, an appropriate action is triggered.
(Etzion, Dori and Nof, 1995, Int’l Journal of CIM)
Purdue-Tivoli Partnership
49
Concluding Thought

“Many companies view each negotiation as a
separate situation, but companies that take a more
coordinated approach are making better deals and
forging stronger relationships”
(Harvard Business Review, May-June, 1999)

Analogy for our research: We can significantly
improve performance by interactions among
participants if an effective, customized protocol
is used to coordinate the interactions needed for
particular environments
Purdue-Tivoli Partnership
50
Purdue-Tivoli Partnership:
Exploiting Purdue’s Technological
Prowess
Arif Ghafoor
Professor, School of ECE
Design and Development of
Distributed Multimedia Systems
Purdue-Tivoli Partnership
51
Long Term Objectives



Design and development of models, tools,
and techniques for multimedia information
management including documents consisting
of text, images, video and audio data
Design and development of internetworking
technology for QoS-sensitive distributed
multimedia applications
Analyzing the impact of multimedia
technology on enterprise IT infrastructure
Purdue-Tivoli Partnership
52
Impact


The research will generate powerful multimedia
information management and workflow models
impacting flexibility, portability, and userconfigurabiliy
The research will provide framework for the
development of heterogeneous multimedia
information systems and applications based on:



workflow models
QoS-sensitive resource management techniques
integration across multiple platforms
Purdue-Tivoli Partnership
53
Impact

The generic Distributed Multimedia System (DMS):




will allow the creation, sharing, management, secure and
efficient access and delivery of multimedia documents
will have potential utility in enterprise-wide IT
infrastructures used for HIS, distributed manufacturing,
business, CSCW environments etc.
Experience with the system will allow to assess the
effectiveness of workflow models, resource
management techniques and network protocols
This research has the potential to significantly
enhance the current state-of-the-art of IT in terms of
developing advanced multimedia applications and
systems
Purdue-Tivoli Partnership
54
Generic Layered Software
Architecture for DMS
Heterogeneity
Multimedia Applications
(workgroup, messaging, interactive)
Distributed Information and
Directory Management
Configuration Management
(broadcasting, multicasting, point-to-point)
OS for End-System Architecture
Network Layer
Purdue-Tivoli Partnership
55
R&D Challenges for DMS
Architecture
Multimedia QoP
Specification
Translation
Network
- End-to-End
Delays
- Jitter Delay
- Bandwidth
- Packet Loss
Rate
Negotiation
Run Time
Resource
Scheduling
Workgroup, Messaging,
Interactive
OS
- CPU Throughput
- Memory Overflow
and Reliability
- Reliability
- Resolution
- Rate of Presentation
- Display Area
- Temporal Synchronization ( Intra/Inter )
Database
- Storage Throughput/
Bandwidth
- Storage Delays
- Distributed
Database
Coordination
Security
- Intrusion Detection
- Access Control
Dependency Model Analysis and QoS Adjustment
System-wide Resource Allocation and Scheduling
Purdue-Tivoli Partnership
56
Proposed Prototype for DMS
Multimedia Applications
(Authoring, Browser, CSCW)
Multiple PC
Platforms
Object Management Presentation Query
Agent
Processor
Agent
Network API
Network (ATM, Fast Ethernet)
Network API
Video DBMS
Video
DB
Oracle DB
...
Network API
Object Manager
Centralized Directory
Text
DB
Purdue-Tivoli Partnership
Image
DB
Audio
DB
Distributed
Multimedia
Servers
(SUN)
57
Research Tasks and Deliverables



Task 1 and Deliverable 1: Development of
networked distributed multimedia database
servers
Task 2 and Deliverable 2: Development of
workflow model(s) and CSCW environment for a
selected application domain (distributed
manufacturing is a potential candidate)
Task 3 and Deliverable 3: Development and
experimentation of different QoS-sensitive
resource management protocols
Purdue-Tivoli Partnership
58
Research Plan




Overall duration of the project: 3 years
Tasks 1 and 2 will commence simultaneously and will
continue for 1.5 years. Task 3 will follow these tasks and
will continue for one year. Experimentation and system
tuning will take six months. Deliverables 1 and 2 will be
submitted after 1.5 year. Deliverable 3 will be submitted
at the end of the project along with the complete software
and final report.
Two GRA’s each year and one PI with 25% AY with two
months summer each year
Approximate Cost: $100K/year
Purdue-Tivoli Partnership
59
Past Research



The PI has established Distributed Multimedia Systems
Laboratory, housed in the School of ECE, in 1992.
A large number of research projects have been sponsored
in the area of multimedia databases, multimedia
networking and distributed multimedia systems by
several government and industrial organizations.
Detail of these projects and research publications can be
found at the following URL:
http://shay.ecn.purdue.edu/~dmultlab
Purdue-Tivoli Partnership
60
Purdue-Tivoli Partnership:
Exploiting Purdue’s Technological
Prowess
Jose Fortes
Professor
Wide-area network computing
Purdue-Tivoli Partnership
61
Long Term Objectives

Demand-driven identification, aggregation,
management, and use of wide-area networked
computing resources




Address usage policy and performance issues
Predictive application-performance modeling
Metacomputing and wide-area distributed
computing
Resource management system for a production
network-computer (PUNCH) used for real-world
applications
Purdue-Tivoli Partnership
62
Impact


Enable sharing of resources (software, hardware,
and personnel) across departments and
institutions
Provide resource management capabilities
crucial to the long-term viability of communityspecific virtual computers enabled by PUNCH
(e.g., DesCArtES, NETCARE)
Purdue-Tivoli Partnership
63
Impact


Allow diverse resource management systems
(e.g., Condor, Globus) to be accessed and used
transparently via a web-accessible virtual
computer
Improve utilization and cost-effectiveness of
high-performance machines by dynamically
adapting resource management policies on the
basis of estimated (predicted) application
performance
Purdue-Tivoli Partnership
64
Research Methodology

Objective:


Demand-driven resource management in a wide-area
networked environment
Challenges:




Usage policies and performance issues intertwined
dynamic, heterogeneous environment; multiple
administrative domains
A priori estimates for resource-usage required
Interoperation with existing resource management
systems
Purdue-Tivoli Partnership
65
Research Methodology

Approach:



Metaprograms; application management
Run-time cost/performance tradeoff decisions
Predictive performance modeling via machine
learning techniques
Purdue-Tivoli Partnership
66
Research Methodology
Purdue-Tivoli Partnership
67
Research Methodology
Purdue-Tivoli Partnership
68
Research Plan

Milestones:







General-purpose performance-modeling system
Core resource-management system design
Language for scheduling queries and usage policies
Deployment in PUNCH; initial evaluation
Scalable, hierarchical architecture for resource
management system
Interoperation protocols for Condor and Globus
Evaluation of system and policies in a production
environment (500+ users and 50,000+ runs per
semester)
Purdue-Tivoli Partnership
69
Research Plan

Deliverables:





Staffing:


Predictive performance-modeling system
Scalable, hierarchical resource management system
Scheduling query and usage policy language module
Interoperability module for external mgmt. Systems
Two graduate students, one research scientist, one
professor
Cost:

$200k/year
Purdue-Tivoli Partnership
70
Past Research

The Purdue University Network-Computing
Hubs -- a WWW-accessible network computer
that allows unmodified software to be used via
standard WWW browsers. To date, 2,000+ users
have generated more than 2 million hits and
have initiated 100,000+ simulations

On the Design of a Demand-Based Network-Computing
System: The Purdue University Network-Computing Hubs.
Nirav H. Kapadia and Jose’ A. B. Fortes. 7th IEEE
International Symposium on High Performance Distributed
Computing. July 1998. Pages 71-80.
Purdue-Tivoli Partnership
71
Past Research

Predictive Application-Performance Modeling in
a Computational Grid Environment -- an initial
prototype of a machine-learning system that
dynamically constructs models for tool resourceusage characteristics with respect to the usersupplied tool input

Predictive Application-Performance Modeling in a
Computational Grid Environment. Nirav H. Kapadia, Jose’ A.
B. Fortes, and Carla E. Brodley. 8th IEEE International
Symposium on High Performance Distributed Computing.
August 1999.
Purdue-Tivoli Partnership
72
Purdue-Tivoli Partnership:
Exploiting Purdue’s Technological
Prowess
Alok R. Chaturvedi Associate Professor
Shailendra Raj Mehta Director, Entrepreneurship Initiative
SEAS - Synthetic Environments for Analysis and
Simulations
Purdue-Tivoli Partnership
73
Long Term Objectives

The objective of our research is to:




Conduct an interdisciplinary program of
research related to business and economic
modeling and simulation
Create realistic virtual representations of
firms, markets and economies
Develop and evaluate computational
models of human behavior
Collaborate with government, industry,
and academic institutions at national and
international levels.
Purdue-Tivoli Partnership
74
Impact


SEAS is a synthetically created economy
with integrated goods and services, stock,
bond, labor, and currency markets.
In these markets two types of agents interact



live: people acting as firms, regulators,
intermediaries.
virtual: artificially intelligent software agents
behave like human agents in a narrow domain.
This environment combines breadth (through
artificial agents) and depth (through human
agents)
Purdue-Tivoli Partnership
75
DoD’s Synthetic Terrain for
War Gaming ..
Air C4I System
Army C4I System
Navy C4I System
USMC C4I System
Mix of Live, Virtual, Constructive Simulations
Units, Platforms, Weapons & Sensors
Purdue-Tivoli Partnership
76
SEAS War Gaming
Environment
Industry System
University System
Government System
Consumer System
Goods &
Services
Stock
Labor
Bond
Currency
Mix of Live, Virtual, Constructive Simulations:
Economies, Markets, Industries, Firms
Purdue-Tivoli Partnership
77
SEAS: Architecture
UPDATE
USER
Decision Enabling Env.
The Internet
QUERY
•Human or
•Intelligent - RESPONSE
Agents
Commercial,
Academic &
Government
Organizations
LAN
Scenarios
REQUEST
Visualization
KNOWLEDGE BASES
RESPONSE
Netcasting
AQUISITION
IMPLEMENTATION
EXTERNAL DATA
•Rueters
•Dow Jones
•IDC
SEAS
Labor Markets
Goods Markets
Financial Markets
Currency Markets
SEAS
SEAS
Labor Markets
Goods Markets
Labor Markets
Goods Markets
Currency Markets
SEAS
Currency Markets
Financial Markets
Financial Markets
Labor Markets
Goods Markets
Currency Markets
Financial Markets
Research Plan


In the next three years we intend to refine the the
computational models to incorporate more complex
human behaviors. We would further develop the global
economy and develop validation methods for large scale
business simulations.
We would request approximately $125,000 a year for
three years. This money will primarily be used to fund a
lab assistant, graduate students, and to acquire data from
companies like IDC, Yankee Group, etc.
Purdue-Tivoli Partnership
79
Past Research

SEAS is an ongoing project. The initial Infrastructure is
already developed through funding from the Department
of Defense, Intel, Ameritech, SAP, and the Institute for
Defense Analyses. We have already developed some
basic computational models of human behavior.

Chaturvedi, A.R., and Mehta, S. R., “Simulations in
Economics and Mangement,” Communications of the
ACM, March 1999
Purdue-Tivoli Partnership
80
Purdue-Tivoli Partnership:
Exploiting Purdue’s Technological
Prowess
Sonia Fahmy
Department of Computer Sciences
Design of Multi-Service Networks
with Multicast Support
Purdue-Tivoli Partnership
81
Long Term Objectives

The research aims at developing protocols
for:


Traffic management for various services
required for multimedia/real-time
applications and bulk data to co-exist
Multicast support for multi-service networks
and feedback control mechanisms (e.g.,
explicit congestion notification and reliable
multicast transport protocols)
Purdue-Tivoli Partnership
82
Impact




The traffic management mechanisms and QoS architecture will
allow multimedia/real-time applications and bulk data to share
the same networking infrastructure, giving throughput and
delay guarantees
Multicasting capabilities will efficiently support collaborative
applications, conferencing applications, distance learning,
searching, and data distribution applications, even for large
numbers of users
The architecture and system developed can affect the services
offered in the future Internet and intranets, improve their
performance and support more complex applications
The simulation tools and prototypes developed will provide a
basis for the deployment of real-time multicast networks and
applications
Purdue-Tivoli Partnership
83
Research Methodology

QoS support:


Simulation tools will be developed and
experiments will be performed to assess the
throughput, fairness, buffer and link
utilization, and packet loss for a set of
carefully selected network configurations
Selective buffer management policies will be
designed and their performance evaluated
Purdue-Tivoli Partnership
84
Research Methodology




End system response to network state
(including explicit feedback) will be designed
and evaluated
Mapping services to network element
behaviors will be designed and evaluated
Control and pricing issues will be
investigated
Prototypes will be built to conduct real
experiments
Purdue-Tivoli Partnership
85
Research Methodology

Multicast support:

Inter-receiver and inter-sender fairness will
be defined within and among multicast
groups
= group
member
Purdue-Tivoli Partnership
86
Research Methodology


Mechanisms for flow/congestion control will
be designed for providing certain levels of
reliability depending on the application type
Services and explicit feedback notification
will be supported for multicast sessions
Purdue-Tivoli Partnership
87
Research Plan

Milestones:





12 months: Buffer management study, response to
feedback study
18 months: Fairness issues for multicasting
24 months: Control and pricing issues, service
mapping study
32 months: Reliable multicast issues
36 months: Multicast extensions to
feedback/services
Purdue-Tivoli Partnership
88
Research Plan

Deliverables:




Architecture, pseudocode and performance studies
for multi-service networks
Architecture, pseudocode and performance studies
for multicast support
Prototype implementations
2-3 graduate research assistants required for 3
years
Purdue-Tivoli Partnership
89
Past Research

ATM-ABR traffic management:



A rate allocation algorithm with delay control
ABR parameter study and the effect of link
bandwidth and round trip time on parameter
values
“The ERICA switch algorithm for ABR traffic
management in ATM networks,” Revised
version submitted to IEEE/ACM Transactions
on Networking, May 1999.
Purdue-Tivoli Partnership
90
Past Research

Multicast support for ATM networks:



A feedback consolidation algorithm and
performance comparison for point-tomultipoint connections
Fairness definitions, a rate allocation and a
feedback regulation mechanism for multipointto-multipoint connections
“Feedback consolidation algorithms for ABR
point-to-multipoint connections in ATM
networks,” IEEE INFOCOM 1998, pp. 10041013, and J. Computer Communications, 1999.
Purdue-Tivoli Partnership
91
Purdue-Tivoli Partnership:
Exploiting Purdue’s Technological
Prowess
Babak Falsafi
Assistant Professor, School of ECE
Impetus: Designing Next-Generation
Distributed Enterprise Servers
Purdue-Tivoli Partnership
92
Long Term Objectives

Distributed Enterprise Server Architectures



Network-Aware Architectures



Shared-Memory Multiprocessors
Server computing: Web, Databases, etc.
Low-Overhead Integrated Network Interfaces
Router Processor/Memory Architectures
Distributed Discrete-Event Simulators

Evaluating Large-Scale Distributed Systems
Purdue-Tivoli Partnership
93
Impact: Distributed Enterprise Servers

Problem





Shared-memory programming, but
non-uniform memory access latencies
Complicates server programmability
e.g., remote-to-local latencies factor of 5~10!
Innovation




Memory access prediction & speculation
Places data next to consumer before access
Remote latency => local latency
Enhances large-scale distributed computing
Purdue-Tivoli Partnership
94
Impact: Network-Aware Architectures

Problem




Architectural innovations do not benefit network
applications!
Static packet processing model
Inflexible network hardware and protocol
Innovation



Network-aware proc/memory architectures
Programmable network interface & routers
Application-specific network protocols
Purdue-Tivoli Partnership
95
Impact: Detailed Machine Simulators

Problem: Simulating large servers




Is both memory and compute intensive
Detailed simulation => 10,000 slowdown
Platform-specific solutions to speed up
Innovation





Fast, accurate, portable simulation
Parallel & distributed simulation on a cluster
Code sampling => reduce workload
Dynamic refinement => model required detail
Direct execution => run at native speeds
Purdue-Tivoli Partnership
96
Research: Big Picture
Server Application
Enterprise Server
Network Protocol
P
P
…..
P
Shared Memory
Distributed Simulator
PC
PC
PC
….
Network Interface
Network Switch/Router
Purdue-Tivoli Partnership
Network
97
Research Methodology

Experimental evaluation



Software




Detailed simulation models
Measurements using H/W monitors
Distributed simulator of large-scale servers
Network switch/router simulators
Application-specific network protocols
Hardware


A cluster of desktops/servers
Fast, high-bandwidth interfaces & switches
Purdue-Tivoli Partnership
98
Research Plan

Request funding




3 students for 3 years
3 years of PI salary for 2 summer months
A cluster of desktops/servers
Deliverables



Designs for simple-to-program large-scale
distributed servers
Designs for network-aware systems
Fast, accurate, & portable distributed simulators
Purdue-Tivoli Partnership
99
Past Research

Enterprise server design





Cost-Effective Parallel Simulation


Reactive NUMA => Sun Microsystems WildFire
Published in ISCA 1997
Speculative Distributed Shared Memory
Published in ISCA 1999
Published in ACM TOMACS, 1997
Network architectures



Application-specific Coherence Protocols
Published in Supercomputing 1994
Coherent Network Interfaces => ISCA 1996
Purdue-Tivoli Partnership
100
Purdue-Tivoli Partnership:
Exploiting Purdue’s Technological
Prowess
Kihong Park
Assistant Professor (CS)
DUNES
Purdue-Tivoli Partnership
101
Long Term Objectives

The objective of my research is:





Efficient/scalable distributed system design
Transparent dependency maintenance
Off-the-shelf
Integrated computation/communication
control
QoS-sensitive scheduling
Purdue-Tivoli Partnership
102
Impact

Development of software system support & tools




Communication-sensitive load balancing library
User-level push/pull caching library
Distributed real-time scheduling library
QoS-sensitive scheduling library


resource contention resolution/arbitration
Prototype System Deployment



DUNES (Distributed UNix ExtenSion)
SBS (Stratified Best-Effort Service)
AFEC (Adaptive Forward Error Correction)
Purdue-Tivoli Partnership
103
Impact (Cont.)

Fundamental understanding of distributed resource
scheduling






Integrated computation/communication control
Distributed real-time scheduling
Integrated network/end system scheduling
QoS-sensitive scheduling
Resource economy
Fault-tolerance, security, and QoS
Purdue-Tivoli Partnership
104
Research Methodology
Top-Down

Top: Modeling & Analysis


Middle: Simulation


framework, qualitative properties
quantitative controlled study
Bottom: System Building

implementation, ultimate test, “buck stops here”
Purdue-Tivoli Partnership
105
Research Plan





Distributed real-time scheduling
Integrated network/end system scheduling
Computational resource economy
Fault-tolerance, security, and QoS
Implement as extensions to DUNES
Purdue-Tivoli Partnership
106
Research Plan (Cont.)


Investigate applicability of existing systems
(e.g., DUNES) to Tivoli environment
Investigate research Tivoli’s research
requirements w.r.t. novelty
Purdue-Tivoli Partnership
107
Past Research

DUNES



NSF ESS-9806741; J. Cruz, C. Gong
Prototype DUNES system (UNIX)
Recent publications


Towards performance-driven system support for
distributed computing in clustered environments,
Journal of Parallel & Distributed Computing, ‘99
DUNES: A performance-oriented system support
environment for dependency maintenance in
workstation networks, IEEE HPDC, ‘99
108
Purdue-Tivoli Partnership
Past Research (Cont.)

Network QoS Architecture



NSF ANI-9875789 (CAREER); S. Chen, H.
Ren
Prototype SBS system
Recent publications


An architecture for noncooperative QoS provision
in many-switch systems, IEEE INFOCOM, ‘99
QoS provision in noncooperative networks with
diverse user requirements, Decision Support
Systems, ‘99
109
Purdue-Tivoli Partnership
Past Research (Cont.)

Multimedia Traffic Control



NSF ANI-9714707; T. Tuan, A. Balakrishnan
Prototype AFEC system for real-time MPEG
video/audio transport
Recent publications


Multiple time scale congestion control for selfsimilar traffic, Performance Evaluation, ‘99
Self-Similar Network Traffic and Performance
Evaluation, Wiley, ‘99
Purdue-Tivoli Partnership
110
For More Information

Network Systems Lab


www.cs.purdue.edu/homes/park/nsl.html
[email protected]
Purdue-Tivoli Partnership
111
Purdue-Tivoli Partnership:
Exploiting Purdue’s Technological
Prowess
Sunil Prabhakar
Assistant Professor
Latency reduction in parallel and
distributed systems.
Purdue-Tivoli Partnership
112
Long Term Objectives

The objective of my research is:


Develop techniques for overcoming network
and I/O latency in distributed systems.
This will be achieved through


efficient data placement techniques
effective replication techniques.
Purdue-Tivoli Partnership
113
Impact


The proposed research will result in the development of
data placement, replication, and access techniques that
significantly improve the performance of distributed
applications.
Due to the high (network and I/O) latency associated
with distributed applications, these improvements will
have a very direct impact on overall system
performance.
Purdue-Tivoli Partnership
114
Research Methodology


The nature of our work is experimental.
We will implement proposed placement
and replication schemes on a collection of
PCs with local disks, connected via
various types of network interfaces such
as 100Mb ethernet, Gigabit ethernet, and
across the internet.
Purdue-Tivoli Partnership
115
Research Methodology


The distribution and placement of data on
the disks will be controlled and we will
populate the system with data from real
applications.
Performance will be measured using this
setup based upon the execution of real
applications.
Purdue-Tivoli Partnership
116
Research Plan




Year 1: identify and analyze a set of distributed
applications to determine their data access
patterns.
Year 2: design and implement alternative
placement and replication schemes
Year 3: evaluate the alternatives
Deliverable: The end product of the project will
be a set of techniques, and tools for their
implementation
Purdue-Tivoli Partnership
117
Research Plan

Resources:



Graduate students
Equipment: PCs, magnetic disks, high speed
local area networks
Estimated budget

$150,000
Purdue-Tivoli Partnership
118
Past Research

Parallel I/O for relational and multimedia
databases.

We have developed state-of-the-art
declustering methods for improved parallel
I/O for range and similarity queries, as well
as for multiresolution image browsing.

“Efficient Disk Allocation for Fast Similarity Searching”, S.
Prabhakar, D. Agrawal and A. El Abbadi. 10th Annual ACM
Symposium on Parallel Algorithms and Architectures (SPAA
98), Puerto Vallarta, Mexico, June 28 - July 2 1998, pages 7887.
Purdue-Tivoli Partnership
119
Past Research

Tertiary storage management

We have developed highly efficient I/O
scheduling algorithms for robotic storage
libraries. We have also developed a novel
mechanism for reliability of tertiary data.

“Tape Group Parity Protection”, T. Johnson and S. Prabhakar.
16th IEEE Mass Storage Systems Symposium MSS'99, San
Diego, California, USA, 15-18 March 1999.
Purdue-Tivoli Partnership
120
Purdue-Tivoli Partnership:
Exploiting Purdue’s Technological
Prowess
Catherine Rosenberg
Associate Professor
Multimedia Networks, Traffic Engineering and
Mobility
Purdue-Tivoli Partnership
121
A Diversified Experience






August 1999 - Present: Associate Professor, Department of
Electrical and Computer Engg., Purdue University. Consultant to
Nortel Networks on Broadband Satellite Network.
January 1998 - July 1999: Head, Department of Broadband
Satellite Networking, Nortel Networks, Harlow, UK. Twice awarded
an Award of Merit. Expert for the European Commission.
Sept. 1996 - Dec. 1997: Head, Department of Traffic Engineering,
Nortel Harlow, UK.
June 1988 - Aug. 1996: Assistant and then Associate Professor,
Department of Electrical Engineering, Ecole Polytechnique de
Montréal, Canada. Consultant to France Telecom.
March 1987 - May 1988: Member of Technical Staff, AT&T Bell
Laboratories, Holmdel, N.J., USA. Twice awarded an Exceptional
Contribution Award.
Oct. 1984 - Oct. 1986: Engineer, ALCATEL, Lannion, France.
Purdue-Tivoli Partnership
122
Long Term Objectives

The objectives of my research are:



Multimedia Broadband Networks (System
Integration, Resource Management,
Technologies)
Traffic Engineering (QoS, Charging,
Dynamic Provisioning, Design, Routing)
IP Mobility
Purdue-Tivoli Partnership
123
Impact and Research Plan

Multimedia Broadband Networks


System Integration: One of the real challenges of
tomorrow’s IT/Telecommunications is in the
integration of technologies for providing seamless
services. This can only be achieved by a ‘system’ team
having an in-depth understanding of the issues at stake
and a vision of the ‘global picture’.
Resource Management: Network Management is of
utmost importance in today’s world of
IT/Telecommunications. At the heart of it is Resource
Management which is key to provide cost-effective
usage of the system while providing QoS. The
Network Management paradigm has to be revisited in
the context of wireless technologies including satellite.
124
Purdue-Tivoli Partnership
Impact and Research Plan

Traffic Engineering



QoS: QoS is key but its introduction in a IP network
will have a tremendous impact on the complexity of the
network management and network dimensioning.
Charging: in a competitive environment where
services cannot be subsidized, there is a need for better
understanding the engineering cost of QoS and GoS as
well as designing tools for dynamic charging to be used
between Network Operators and ISPs.
Design and Dynamic Provisioning: Tools are needed
to measure the right performances and take automatic
decisions to provision IP networks. This requires an indepth knowledge of IP networks and their design and
routing.
125
Purdue-Tivoli Partnership
Impact and Research Plan

Horizontal Integration for IP Mobility


The question is: How to offer seamless IP mobility
everywhere (indoor/outdoor, urban/rural,
nationwide/global coverage) ?
The answer being: By taking advantage of the full
range of technologies in an integrated way (wireless,
satellite, terrestrial) by combining micro mobility with
macro mobility and managing them in an integrated
and efficient manner.
Purdue-Tivoli Partnership
126
Research Methodology






Team work, close links with industry and
standardization awareness.
Starting from a detailed and precise problem
formulation to achieve a clear understanding of
the solution space and create the most appropriate
set of solutions.
Innovative system/protocol/algorithm design.
Use of analysis as much as possible.
Use of simulation to assess performance impact of
solutions.
Development of testbed if appropriate.
Purdue-Tivoli Partnership
127
Past Research: Papers

A. Girard, A. Meddeb and C.R; Design of Broadband Networks with
Multipoint Connections, IEEE/ACM Trans. on Networking, submitted 7/99.
 C.R, End-to-End Resource Management Integrating Multiple Access,
Bandwidth on Demand and Call Admission Control for ATM Geostationary
Satellite Systems, IEEE Com Magazine, submitted 4/99.
 A. Conway and C.R; Weighted Fair Blocking Mechanisms, Performance
Evaluation, submitted 3/99.
 S. Delas, R. Mazumdar and C.R, Cell Loss Asymptotics for Finite Buffers
with HOL Priorities, IEEE/ACM Trans. on Networking, submitted 5/99.
 H. Yaiche, R. Mazumdar and C.R; A Game Theoretic Framework for Rate
Allocation and Charging of Elastic Connections in Broadband Networks;
IEEE/ACM Trans. on Networking, submitted 12/98.
 R. Mauger and C.R; QoS Guarantees for Multimedia Services on a TDMAbased Satellite Network, IEEE Com. Magazine, 7/97.
 S. Ramesh, C.R and A. Kumar; Revenue Maximization in ATM Networks
Using the CLP Capability and Buffer Priority Management, IEEE Trans. on
Networking, 12/96.
Purdue-Tivoli Partnership
128
Past Research: Patents
Title
Connection Oriented Routing
Connectionless Communications
Network
Connectionless Network
Integrated Connection Admission
Control And Bandwidh On Demand
For Multiple Access ATM Like
Network
Integrated Signalling for ATM like
Networks
Satellite Communications Routing
Algorithm
Fairness and Aggregation in
Telecoms Networks
West Early Bird
SpaceWEB
Method and Apparatus for
Distributed, Hierarchical Satellite
Network Control
Status/Date
Disclosure Filed, 18/4/97
Disclosure Filed 1/8/97
Inventors
CR et al
CR et al
Disclosure Filed 2/2/98
Disclosure Filed 16/7/98
CR
CR et al
Disclosure Filed 16/7/98
CR et al
Disclosure Filed 11/1/99
CR et al
Disclosure Filed 30/6/99
CR
Disclosure Filed in France
Disclosure Filed 7/99
Disclosure Filed 7/99
CR et al
CR et al
CR et al
Purdue-Tivoli Partnership
129
Purdue-Tivoli Partnership:
Exploiting Purdue’s Technological
Prowess
Aditya P. Mathur
Professor
Testing, Monitoring, and Controlling
CORBA-based Distributed Systems
Purdue-Tivoli Partnership
130
Long Term Objectives



Provide a commercially viable methodology
for testing CORBA compliant applications.
Provide a methodology for monitoring and
control of CORBA compliant applications.
Provide a tool to support the above
methodologies.
Purdue-Tivoli Partnership
131
Impact (1)



TDS 1.1, developed at Purdue, will allow
measurement of test adequacy for components and
systems and assist in assuring high quality
distributed applications. TDS is the only tool of its
kind available today.
Tivoli, BT Labs (UK), and Telcordia have indicated
their willingness to use TDS 1.1 in ongoing projects.
TDS 2.0 will allow distributed monitoring and
control of distributed applications. TDS 2.0 is
expected to be available for use in May 2000.
Purdue-Tivoli Partnership
132
Impact (2)


Our research has the potential to create new jobs in
the state of Indiana. A start-up company is
considering commercialization of TDS 1.1 and its
forthcoming versions.
Availability of TDS 2.0 will provide a unique
opportunity to entrepreneurs in the state of Indiana
to set up commercial ventures such as Distributed
Systems Test and Monitoring Laboratory, On-line
Software Rental House, Component Quality
Assurance, and Commercial Data Bank.
Purdue-Tivoli Partnership
133
Research Methodology (1)



Develop new criteria to evaluate the
adequacy of testing: Completed in May 98.
Experimentally evaluate the adequacy
criteria: Scheduled for completion in December 99.
Collaboration with Tivoli.
Develop a methodology for monitoring and
control: Completed July 99. Collaboration with
Telcordia and Tivoli.
Purdue-Tivoli Partnership
134
Research Methodology (2)

Develop a method for the dynamic extraction
of application architecture: Scheduled for
completion in December 99. Collaboration with
Telcordia and Tivoli.

Develop a commercially usable tool that
incorporates the above criteria and
methodologies: Version 1.1 available in August
99, Version 2.0 available in May 2000.
Purdue-Tivoli Partnership
135
Research Plan (1)



Implement a monitoring and control
methodology: August 99-December 99.
Implement dynamic extraction of architecture:
August 99-December 99.
Integrate the above into TDS 1.1, evaluate
performance, refine/tune tool, and release TDS
2.0: January 00-May 00.
Purdue-Tivoli Partnership
136
Research Plan (2)



Develop requirements coverage criteria,
implement in TDS 3.0, evaluate it
experimentally: August 00-December 00.
Develop, implement, and evaluate tracing and
test execution in TDS 3.0: August 00-December 00.
Prepare TDS 3.0 release: January 01-May 01.
Purdue-Tivoli Partnership
137
Budget


Staffing, Equipment, and Software:

Graduate students: 4

Programmer: 1

PI: 1

Equipment and software
Estimated cost: $400K over 2 years.
Purdue-Tivoli Partnership
138
Past Research




Experimental evaluation of white-box coverage
criteria. Sponsors: NSF, Telcordia, SERC.
New methods for the estimation of software
reliability. Sponsors: NSF, Telcordia, SERC.
Listen Project to investigate the use of sounds in
program monitoring. Sponsor: NSF.
Use of parallel machines in testing software for
ultra high quality. Sponsor: SERC, NSF.
Purdue-Tivoli Partnership
139
Past Research



Over 70 publications have resulted from the
above research. Technical reports are available
in Technical Report Reading Room at:
http://www.cs.purdue.edu/serc
Sponsors have invested over $2M over 10 years
in our research during 1988-1998.
Lessons learned from past research are helping
us direct the course of the ongoing research.
Purdue-Tivoli Partnership
140
Purdue-Tivoli Partnership:
Exploiting Purdue’s Technological
Prowess
Jens Palsberg
Associate Professor
Software Security in Distributed Systems
Purdue-Tivoli Partnership
141
Long Term Research Objectives



Build safe, secure, and highly-optimized
mobile code.
Prevent mobile programs from leaking a
host computer’s secrets.
Protect the intellectual property contained
in mobile code.
Purdue-Tivoli Partnership
142
Potential Impact




Safer web-based commerce.
More secure wireless connections for
laptop computers.
Safer dynamically-configurable missioncritical networks.
Mobility of platform-independent code
without compromise of proprietary
secrets.
Purdue-Tivoli Partnership
143
Research Directions (I)

Typed Assembly Languages for compiling
Java: to address security in low-level code.


Can we optimize method calls to improve
execution speed while retaining guarantees of
memory safety?
Can we have effective type-checking while
retaining the expressiveness necessary to
write useful programs?
Purdue-Tivoli Partnership
144
Research Directions (II)

Secure Information Flow Model: to prevent
the leaking of secrets.


Can we formulate a useful confinement
property for distributed and concurrent
systems?
Can we devise an effective way of checking
such a confinement property?
Purdue-Tivoli Partnership
145
Research Directions (III)

Software Obfuscation and Watermarking:
to protect intellectual property.


Can we make a quantitative and qualitative
assessment of current software obfuscation
techniques and tools?
Can we embed a watermark in a Java
program which is resilient to the standard
attacks?
Purdue-Tivoli Partnership
146
Research Methodology


Our methodology is mostly experimental,
and based upon earlier theoretical work.
Completed preliminary experiments:


A Typed Assembly Language for Java is up
and running. The next version is underway.
Two watermarking systems for Java 1.1 have
been completed. Large-scale experiments are
in progress.
Purdue-Tivoli Partnership
147
Research Plan

Milestones:





A certifying compiler from Java to a Typed
Assembly Language.
A confinement checker for concurrent and
distributed processes.
An efficient and resilient Watermarking System for
Java.
Staffing: 1professor, 1 post-doc, 5 students.
Budget: $150,000 per year.
Purdue-Tivoli Partnership
148
Technical Reports (Summer 1999)



Dennis Brylow and Jens Palsberg. “A
Typed Assembly Language for Java.”
Sowmya Krishnaswamy, Minseok Kwon,
Di Ma, Jens Palsberg, Qiuyun Shao,
Christina Yi Zhang. “Experience with
Software Watermarking for Java.”
Current Funding: NSF, IBM, CERIAS, etc.
Purdue-Tivoli Partnership
149
Purdue-Tivoli Partnership:
Exploiting Purdue’s Technological
Prowess
Jan Vitek
Assistant Professor
SECURE MOBILE OBJECT SYSTEMS
Purdue-Tivoli Partnership
150
Long term goals of the research:

Technologies for engineering distributed
objects applications



adaptive
location-aware
high-assurance
Purdue-Tivoli Partnership
151
Impact
Facilitate the construction of large distributed
information systems, e.g. workflow management in a
large corporations.
These systems are designed and implemented in a
decentralized fashion, but some enterprise wide security
policies must be enforced.
Leverage object-oriented principles of reusability to
reduce design and implementation effort.
Purdue-Tivoli Partnership
152
Research Methodology

Sound formal basis

Security properties must be studied in a
formal setting.




Concurrent process calculi and notions of
behavioral equivalence
Identify relevant security properties
Properties preserved under composition
Widely applicable language mechanisms for
enforcement of security
Purdue-Tivoli Partnership
153
Research Methodology

... backed by real systems

SecureJava: Secure Composition of Untrusted Code


High-assurance Java for composing untrusted or
partially trusted classes. Type systems and
compositional security properties.
FOAM: Featherweight Objects and Agent Mobility

Compact object and byte code formats and data
compression tool integrated with RMI/JINI.
Lightweight virtual machine design.
Purdue-Tivoli Partnership
154
Research Methodology

... backed by real systems

DIGIDOC: Digital Protection for Active Electronic
Documents
 Distributed infrastructure for the exchange and
protection of active documents (workflow)
incorporating protection mechanisms for policy
enforcement.
Purdue-Tivoli Partnership
155
Past Research

JavaSeal
Secure mobile agent system implemented in Java.
Support for the delivery of self-contained distributed
application. [10 man years, Swiss SPP]
Vitek and Bryce, “The JavaSeal Mobile Agent Kernel”,
In Agent Systems and Application’99.

HyperNews
Commercial digital content delivery system built
with JavaSeal.
Konstantas, Morin, Vitek, “MEDIA: A Platform for the
commercialization of electronic documents.
Purdue-Tivoli Partnership
156
Past Research

JAZZ
High-density Java bytecode format and
compression software.
Horspool, Bradley, Vitek, “JAZZ: Tailored Compression
of Java bytecode”, IBM CASCON’98.

SPL
Secure object-oriented programming
technology.
ECOOP’98, OOPSLA’99, CSFW’99
Purdue-Tivoli Partnership
157
Past Research

SEAL
Programming language for mobile
computation.
WIPL’97, JFLA’99, ERSAD’97

Language implementation and static
analysis
OOPSLA’97, ECOOP’97, JLMC’97, CC’97, ECOOP’95,
ECOOP’94, CC’92, ICCL’92
Purdue-Tivoli Partnership
158
Purdue-Tivoli Partnership:
Exploiting Purdue’s Technological
Prowess
Bharat Bhargava and Sanjay Madria
Experimental Studies in Adaptable
Distributed System Software
www.cs.purdue.edu/faculty/bb
Purdue-Tivoli Partnership
159
Long Term Objectives



Investigate and develop adaptable distributed system
architecture/implementation and conduct
experiments that provide a single point of
management and control for the myriad of
interconnected components; applications, software,
middleware, databases, and evolutionary platforms
Investigate adaptable replica management schemes
for version control, change management, consistent
updates of software and data files
Scalability experiments - horizontal (geographical
distribution) and vertical (number of sites)
Purdue-Tivoli Partnership
160



Experimental studies - network behavior,
reliability, interoperability , integration, and
monitoring. Impact on quality of service (QoS)
in emerging distributed applications such as ecommerce
Software support for data compression ,
checkpoint/restart (recovery) and user
authentication
Solutions embedded in ESM and Galileo at
Tivoli
Purdue-Tivoli Partnership
161
Impact



Our research will impact in development, evaluation,
and experiments with enterprise support management
(ESM) product for asset, change and network system
management. Specifically, we will contribute to expert
evaluation, expert views and expert mail agent in ESM.
Our research on nested and workflow transactions will
contribute to the efforts of Galileo at Tivoli. The
integration of business process automation with other
Tivoli products will be enhanced.
Experience with collaborative software development
and information systems reliability mechanism to deal
with various types of failures (RAID)
Purdue-Tivoli Partnership
162



Our results are applicable to systems software used in
electronic commerce transactions, trading of stocks,
collaborative software, distributed file systems,
multimedia/video conferencing.
Our system will allow non-stop access to a customer
to place an urgent order even if network router is
down or sites have failed. We assume that a single
transaction must travel across multiple unsecured
networks that are not visible to network administrator
or customer
Automatically distribute updates of software or data
across the multiple sites. Good for view maintenance
in case of multiple business partners are involved.
Purdue-Tivoli Partnership
163


Centrally manage and monitor all network,
application and system activity and adapt to
various types, extent, and duration of anomalies
Replication and version management of software
under various anomalies to provide optimum
availability
Purdue-Tivoli Partnership
164
Research Methodology



We develop, implement and experiment with
algorithms, models and system software tools (Raid,
O-Raid, P-Raid, MM-Raid, AVC, WANCE, active
network engine, etc).
We build prototype systems and software tools,
develop bench-marks to evaluate distributed software
performance/capability and transition to commercial
grade embedded software
We conduct a series of experiments over LAN and
WAN using sites across the world using WANCE tool
Purdue-Tivoli Partnership
165
Sample Experiments

Replicated copy control (version management)
experiment - to study the effect of copier transactions and
mechanisms for refreshing out-of-date copies. Examine effects
of partial replication. Study availability vs degree of replication
using Raid system and TP benchmark

Adaptability/Embedding experiment





Identify cost of reconfiguration and adaptation.
Is throughput reduced during adaptation?
Overheads due to Embedding
Vary failure parameters (frequency, duration, type, extent,
timing)
Measure cost of adaptation and benefits in terms of
availability, response time, etc.
166
Purdue-Tivoli Partnership
Research Plan




Extend embedded capabilities in distributed system
(Raid) and conduct experiments to deal with
adaptability in update management (replica control) and
remote software management.
Algorithms to deal with site failures, network
partitioning, and security.
Measurement of reliable communication over
communication software for replica management
(change management), non-stop availability and
transmission of large software objects over WAN/LAN
Operating system changes for improving
communication software support for embedded systems
Purdue-Tivoli Partnership
167





Performance of adaptability and reliability
schemes to deal with failures, constraints in
system resources, heterogeneity and
communication (LAN, WAN, MAN, Mobile)
Deal with various traffic patterns of transactions,
multiprogramming level, scalability to number of
sites and distance among sites (WANCE tools)
Experiment with distributed, collaborative and
cooperative execution of business transactions
Integration in the framework of ESM and Galileo
Development of policies for adaptability at the
application, system and network layer to meet
varying QoS requirements.
Purdue-Tivoli Partnership
168
Deliverables




Guidelines for change management system
adaptability, cost-benefits analysis and evaluation.
(June 2000)
Prototype implementation and experiments with
technology transfer for various components of
distributed software (ESM) (Dec. 2000)
Software Tools (ANE, WANCE, mini-Raid) for
experiments in scalability, wide area network,
reliability, active networking and security (June
2001)
Recovery and adaptation to failures of various types,
duration, extent, timing in context of embedded ESM
and Galileo (June 2001)
Purdue-Tivoli Partnership
169

Past Research/Accomplishments
Awards -






Best Paper Award on Adaptability in Distributed
Systems
IEEE Fellow Award
IEEE Technical Achievement Award for work on
Distributed Systems
14 Ph.D., 20 MS and 15 undergraduates have
worked on our projects.
Projects (and student fellowships) are supported by
various funding agencies NSF, Army, NASA
including industries like IBM, AT&T, Unisys etc.
IBM supported project - Integrated Analysis and
testing of Distributed Software
170
Purdue-Tivoli Partnership


RAID Distributed System - Reliability,
Adaptability, Interoperatibility, Distributed
Systems - IEEE Transactions on Software
Engineering
Reliability - models, performance,
implementation (book - Advanced Replication
Techniques for Distributed Systems, 1996)

Integrity - TSE, TKDE’99

Site failures- TKDE’99
Network partitioning- Computer Journal’98,
Recovery- DEXA’97
Checkpointing- SRDS’98, TKDE’99



Purdue-Tivoli Partnership
171

Adaptability






Failures (VLDB)
Formal models - IEEE TKDE’99, Information
Systems,97
Adaptable communication software - ASSET’98,
multimedia’96,
Adaptable commit - IEEE RQD’99
Dealing with multiple heterogeneous data,
algorithms, data models - Journal of Multimedia
Systems’99
Interoperability
 Degree of autonomy
 Degree of consistency- IEEE TKDE’99
Purdue-Tivoli Partnership
 CORBA
172

Distributed systems
 Transaction processing - TKDE,99
 Operating system support
 Communication software - Multimedia, 99
 World Wide Web data management (Sp. Issue
of WWW journal), DASFAA’99, FODO’98,
IDEAS’99
Purdue-Tivoli Partnership
173
Budget/staff


Staff - Bharat Bhargava, Sanjay Madria, five
graduate students and three undergraduates
Budget - Direct costs $150,000 per year for 3
years - total $ 450,000
Purdue-Tivoli Partnership
174