Java on Wall Street

1
Java on Wall Street

• Evolving the J2EE presentation layer

Patrick J. Smith
Guy Smith
2
Java Rich Clients

• What are Rich Clients and why do I care?
• Drivers
• X-internet
• Technologies SWT, Flash, Faster Swing
• Communication - SOAP
• Companies deploying rich clients
• What do they have in common
• What are the management issues
• Deutsche Banks - DB Ticket Architecture
• Deutsche Banks Client Provisioning Strategy

3
B2C classic architecture
Database Layer
Client Layer
Web Layer
Oracle PL/SQL
FAT Client
Load Balancer
Mainframe
Application Server
iPlanet
Region 1
Browser Client
iPlanet
Region 2
DB2
4
Logic Shift from Legacy J2EE
Database Layer
Web Layer
Server Layer
Client Layer
Application Server
Browser Client
Browser Client
Oracle PL/SQL
App Server
Web Server
Browser Client
Browser Client
Load Balancer
Mainframe
Browser Client
Browser Client
Application Server
Web Server
Browser Client
Browser Client
App Server
Region 1
Web Server
Browser Client
Browser Client
Region 2
DB/2
Browser Client
Browser Client
5
Complex browser applications - Performance
Database Layer
Web Layer
Server Layer
Client Layer
Application Server
Oracle PL/SQL
App Server
Web Server
Browser Client
Mainframe
Application Server
Web Server
App Server
Region 1
Web Server
Region 2
DB/2

• Round trips to the server impede performance
• Limited inter-dependant windowing capability
• Complex state engines not feasible

6
Complex applications Complex Maintenance
Database Layer
Web Layer
Server Layer
Client Layer
Application Server
Oracle PL/SQL
Page 1
JSP
JSP
App Server
Page 2
JSP
Mainframe
JSP
JSP
Application Server
Page 3
JSP
App Server
Region 1
Page n
JSP
JSP
Region 2
DB/2

• JSP leads to high degree of coupling for complex
  applications
• High cost of maintenance
• Increase in total cost of ownership

7
Rich Clients start emerging as J2EE presentation
layer alternative in 2000
Database Layer
Web Layer
Server Layer
Client Layer
Application Server
Browser Client
Browser Client
Oracle PL/SQL
App Server
Web Server
Browser Client
Browser Client
Load Balancer
Mainframe
Application Server
Web Server
App Server
Region 1
Rich Client
Rich Client
Web Server
Region 2
Rich Client
Rich Client
DB/2

• Offload server performance gains
• Rules driven workflow complex GUIs
• Thin on logic but heavy footprint

8
Who is using this architecture?
9
Rich Client Architecture Model View Controller
Client Layer
Client Layer
View
Server Layer
Controller
Model
Web Server
HTTP-s
App Server
Oracle PL/SQL
10
Client Layer
Client Layer
View
View
View
View-Controller
View-Controller
View-Controller
Model
Navigation Controller
Navigation Controller
Navigation Controller
HTTP
E
Event Triggers
E
Server Layer

• Navigation Controllers and Views
• Windowing navigation state engine
• Provides application workflow control
• De-coupled from individual view controllers
• Presentation Layer communicates via CORBA/SOAP

Web Server
App Server
Oracle PL/SQL
11
Rich Clients introduced new management challenges
Database Layer
Web Layer
Server Layer
Client Layer
Application Server
Oracle PL/SQL
App Server
Web Server
additional files
jar
jar
jar
Load Balancer
Mainframe
JRE 1.3.1
Application Server
Rich Client v.1
Web Server
App Server
Region 1
Web Server
Region 2
additional files
DB/2
jar
jar
jar
JRE 1.4..1

• Initial deployment
• Provisioning incremental business functionality
• Rollout/Rollback
• Version and access control
• Fault tolerance, monitoring and auditing
• JRE management
• High Rev/High Capability Xtreme/CMM

Rich Client v.2
12
Key issues

• High availability java application management
  solution
• Mitigate risk of application failure
• Critical error detection rollback accelerates
  time to repair
• Accelerate provisioning of critical business
  functionality to end users

13
Java on Wall Street

• DBTicket Architecture and Application Management
  Approach

Patrick J. Smith
Guy Smith
14
Deutsche Bank - DBTicket STP Architecture
Database Layer
Corba Layer
Server Layer
Client Layer
Application Server
SYBASE
Trade Capture Server
London
OS Agent
Frankfurt
Trade Browser Server
SYBASE
Name Service
New York
Workflow Server
Madrid
Event Service
Milan
SOAP Server
Hong Kong
Deployment Server
Tokyo
Sydney
15
Deutsche Bank - DBTicket STP ArchitectureUsing
agent-based deployment to a geographically wide
base
Database Layer
Corba Layer
Server Layer
Client Layer
Application Server
SYBASE
Trade Capture Server
A
OS Agent
files
jar
jar
jar
JRE
SYBASE
Trade Browser Server
Rich Client v.1
Name Service
Event Service
Workflow Server
A
files
jar
jar
jar
JRE
Rich Client v.2
Deployment Server
Fixed Servers
User Base
16
Multiple environments, one admin experience
DEV
TST2
17
Server Layer
Client Layer
Client Layer
Build System
Rich Client
A
Version 1.0
Recent Build
Java Deployment Server
Differencing Engine
Version Vault

• Build System Integration
• New builds are uploaded to deployment server
• Command line scripts or Admin console

18
Server Layer
Client Layer
Client Layer
Build System
.TXT
A
.DLL
Version 1.0
.EXE
JARs
JRE
Java Deployment Server
Differencing Engine
Version Vault

• Client-side Agent On-demand provisioning
• User starts the java client
• Agent intercepts startup and requests
  authorization and access
• Server calculates class differences between jar
  files

19
Server Layer
Client Layer
Client Layer
Build System
.TXT
A
.DLL
.DLL
Version 2.0
.EXE
JARs
JARs
JRE
JRE
Java Deployment Server
Differencing Engine
Version Vault

• Client-side Agent Provisioning new features
• Differences are delivered to client agent
• Agent physically updates ALL JARS and files
• Audit statistics recorded on the server

20
Server Layer
Client Layer
Client Layer
Build System
.TXT
A
.DLL
Version 2.0
.EXE
JARs
JARs
JRE
Java Deployment Server
Differencing Engine
Version Vault

• Client-side Agent Fault Tolerance
• Unhandled exceptions caught by the agent
• Agent alerts server of production service outage

21
Server Layer
Client Layer
Client Layer
Build System
.TXT
A
.DLL
.DLL
Version 1.0
.EXE
JARs
JARs
JRE
JRE
Java Deployment Server
Differencing Engine
Version Vault

• Client-side Agent - rollback
• Cached differences are delivered
• Agent updates client
• Production uptime is maintained

22
Single Agent Strategy
Multi-Agent Strategy
Client Layer
Client Layer
Rich Client
Server Layer
Server Layer
Application Server
Application Server
Rich Client
Deployment Server
Deployment Server
A
Rich Client
Rich Client

• Smaller footprint
• Clients are stored in cache
• Single point of failure

• Larger footprint
• Redundancy, no single point of failure
• Allows selective update of agent

23
Thank You
Patrick J. Smith
Guy Smith