An Architecture Proposal for Enterprise Message Brokers Using Messages to link components and hosts

1
An Architecture Proposal for Enterprise Message
BrokersUsing Messages to link components and
hosts

• Jörn Guy SüßComputergestützte Informationssysteme
  CISTechnische Universität Berlin
• Michael MewesFraunhofer-Institut für Software-
  und Systemtechnik ISST
• November 2nd, 2000

2
Overview

• The field
• Why messaging?
• Why brokers?
• Why broker methods?
• Why a reference architecture proposal?
• The architecture
• Design goals
• Nodes and Queues
• Transformation Engine
• Transformation Engine Features
• Advantages of the Architecture
• Conclusion Outlook

3
A typical business scenario
Mail SC
K44

• The customer wants...
• No mistakes (reliable transactions )
• No changes (loose coupling)
• low price
• quick!

4
Synchronous does not work

• because we have ...
• Different data types
• Different network protocols
• Different binaries
• No reusable interfaces

5
More reasons why Synchronous does not work

• because we have ...
• Orbs on hosts are buggy complex
• Orbs create security problems
• Orbs on hosts are expensive
• Projects are slow and difficult

6
Asynchronous does work!
Local TX
Local TX
Mail SC
K44

• because we have...
• ASCII128
• a report writer/ database dumper
• files
• ? We can create XML docs !
• local transactions

7
So, where is the problem?
ERPSC
N² stovepipe interfaces No funds for big
vision No plan for big vision
K44
Mail SC
PDM
ERP2
We need a method and a tool to tackle the message
chaos!
8
What do we need to solve it?

• ? MethodUnified Process for Messaging

9
What do we need to solve it?

• ? MethodUnified Process for Messaging
• ? LanguageXML, UML (Use Cases!)

10
What do we need to solve it?
? MethodUnified Process for Messaging ?
LanguageXML, UML (Use Cases!) ? Tool message
transformer (the message broker)
MBMs research program will cover all these
items. This talk is about item ?.
11
The Broker Scenario
Creating a structured message Reliable delivery
to the broker Transformation of the
message Reliable delivery to the
subscribers Deciphering a structured message
Creating a structured message
Creating a structured message Reliable delivery
to the broker
Creating a structured message Reliable delivery
to the broker Transformation of the message
Creating a structured message Reliable delivery
to the broker Transformation of the
message Reliable delivery to the subscribers
12
MB-Component Shopping
IBM MQ,BEA TUX,TIB/Rendezvous MS MQ
DaimlerChrysler,MICO Orb
?
IBM MQ,BEA TUX,TIB/Rendezvous MS MQ
DaimlerChrysler,MICO Orb
Compile
Build
Build
Node 2
Node 1
13
Reference Architecture ComponentsSubmission and
Retrieval Adapters

• link custom synchronous systems
• architecture description
• interface/mesg. description
• platform compiler
• also for retrieval

DCD/DTD
Build
Node 1
DaimlerChrysler Adapter
14
Background DaimlerChrysler Language

• Creator Klaus Wissmann
• Context Diploma thesis at the University of Ulm
• Title Erweiterte Konzepte der Funktionsintegratio
  n 7/2000
• Content XML-based language FIX describes

15
Testing Message Brokers

• Qualitative Test MM99
• Business scenario examination
• Established Enterprise Critical Criteria
• First functionality tests
• Aftermath Product deficiencies!
• Quantitative Test EE01
• Performance Benchmark
• Guidelines to establish ECC-fulfillment
  numerically
• Products tested
• TIBCO MB,MSI Mercator, MS BizTalk
• No participation
• IBM MQSI V2 Complex, Offered only with
  consulting
• STC eGate Evolved, Industrial leader

16
Test Results
17
Our design goals

• Simplicity
• Scalability
• Design Reuse
• Use of Standards / Openness
• Access to Components / Platforms

18
Nodes and Queues
??
??
??
??
Node 2
Node1/?
Node 2/?
Node 1
??
??

• Nodes are only represented by their published and
  subscribed DTDs

• Nodes can be supervised by the amplitude of their
  queues.

• Nodes may attach to / disconnect from the engine
  as long as queues are empty

19
Transformation Engine - like JSP !
Declarative description
HTML ( Script)
Demand driven compilation
Compile
High-performance executable
20
Transformation Engine Activation Policy
21
TE Feature Micro Workflow
Mail SC books container space K44 has to reserve
the containers ERP2 has to provide the personal
Overhead avoidable? Yes! - Auto-Queue!
22
Engine Trouble Cascading Transformation I
?
?
??
?
?
?
??
??

• Recursion is dangerous
• must run occurs check on insert
• occurs check is NP-complete!
• remedy queue-length-tracking to assist operator

23
Engine Trouble Cascading Transform. II

• Wasteful cycles
• dead loops no subscribers
• fruit loops spin-off documents flood queues

• Detecting cycles
• track deviation from queue length statistics
  (variance, mean)
• operator can trace backlog
• operator must cut cycle manually

24
TE Feature Component Binding

• Cant I have the VAT-calculation, part-Number
  lookup, I_NEED_IT_DAILY-function in the
  TRANSLET?

• YES! XSLT has hooks for middleware
• But
• XSLT-Docs will be proprietary
• No data types
• No error handling, just logging
• Transformations not repeatable
• Time-Out Required for state management

25
Load Balancing and Platform Independence

• Design reuse in different project scopes

26
Defeating Features

• Matching
• Describe matches by rules
• Handle Transactions/States
• Handle load balancing
• Workflow
• increases complexity
• not well-understood
• dead-lock conditions

27
Advantages of the Architecture
Openness, Simplicity XSLT Scalability
JSP-Design Access to Components XSLT-Extion
Functs
28
Conclusion Outlook

• Method
• Required for broker to be useful
• Requires a platform architecture with minimum
  properties
• Should be based on UML
• We will derive, apply and document the method
  within the next 2 years
• Architecture
• proposed architecture offers the minimum
  properties
• It is open, scalable, simple and (re-)useful
• We expect similar products/implementations within
  the coming year.

29
(No Transcript)