CS 4448 - Spring 1998
Object-Oriented Programming and Design
Talk 11.2.2
by
Joshua Given

Paradigm Shift

• Xerox PARC saw three distinct phases of computing:
  1. Host-Centric (mainframes) - Expensive computing, many users interact with a single machine
  2. Desktop-Centric (PC's) - Computers on every desktop, connected by LAN
  3. Network-Centric (NC's) - Computing power cheap and globally connected
• The transition to this network-centric paradigm will change software development drastically.
• "Software development, deployment, and maintainance must evolve to deal with the new realities."

Three-tier Architectures

• Software Development will switch from a client/server to a three-tier architecture (Network Computing Architecture):

      Web Client  <-Message->  Business Object  <-DB access->  Data Persistence
      

  • Web Client - renders data in the form of HTML pages and executes behavior in the form of Java, JavaScript, or ActiveX/DNA controls. It handles all aspects of presentation and has little knowledge of the business rules of an organization.
  • Business Object - implements the presentation and data storage-independent logic of the disributed application
  • Data Persistence - Persistence stores are file systems, relational and object DB's
  • Message mechanisms include RPC, HTTP, IIOP, RMI, and COM+
• The first widely-deployed NCA appeared in 1995 when Web servers added CGI and DB vendors provided Perl API's. Clients requested pages that where constructed by Perl CGI programs retrieving data from DB's and handling forms input.
• Companies and organizations are competing heavily in the architecture space.
• Advances in browsers, Java, Web servers, DB's, and transactions servers are happening because the new architecture solves many of the difficulties of network computing -
  • deployment of data and behavior - clients can download HTML pages containing Java or JavaScript
  • security - Java and JavaScript provide an extremely high level of security for clients
  • firewall tunnelling - email and HTTP have tunnelling, many protocols layer or tunnel through HTTP
  • performance and scalability -
    • thin clients to be widely deployed, can concentrate on presentation logic.
    • Business logic moved to an object on a server.
    • Distributed applications can have more clients than client/server apps
    • Thick servers optimized for computation or throughput, load-balanced across different machines and made fault-tolerant

Vision

• Need for components and distributed objects
• Benefits of objects: Encapsulation, inheritence, and poplymorphism.
• Components provide design-time and runtime services for software objects:
  • object methods
  • event and property discovery and invocation
  • persistence
  • licensing
  • ....
• Distributed objects - object that are distributed in space
• Componenets can invoke methods on remote objects almost as easily as a local object method invocation. This extends RPC.

Distributed Object Comparison

• In reality, all distributed object technologies are roughly similar, with specific differences dependent upon the aim and legacy of the technology.
• All provide for compiletime and runtime discovery and use of properties, methods, and events of remote objects.
• There are diferences, however:

  	Java RMI	COM	IIOP
  Development Language	Java	C++,VB,Java	Any
  Platform	Many, not Mac	Win x86, 95 & NT	Many
  Standard	Closed	Quasi-open	Open
  Object-Oriented	Yes	Quasi - no inheritence	Yes
  Remote interface definition	Java	MS IDL, Object Definition Language (ODL)	Java, Interface Definition Language (IDL)
  Unique Objects	Quasi	No	Yes
  Ease of development	Very easy	Hard	Easy
  Scalability	None	MTS	Emerging

  • Scalability refers to the numbers of clients and the number of server objects supported.

Benchmark

• A simple banking app was created by the author using a variety of protocols
• Run on a single Pentium 133 NT Workstation ThinkPad
• Trivial case of an integer amount being deposited in the bank

  Access type	Average time
  Procedure calls	.01 ms
  Sockets - connected	3 ms
  HTTP/CGI	650 ms
  HTTP/Servlet	100 ms
  RMI - connected	10 ms
  RMI - connectionless	50 ms
  IIOP - connected	6 ms
  IIOP - connectinless	50 ms
  COM	1 ms
  DCOM	???

Caveats

• Many issues were not addressed in benchmark
• Scalability of app is unknown
• Another issue is start-up performance - varying levels of code must be downloaded
• Understanding the style of the app's behavior is crucial

Components

• Beans and COM are similar, but there are diferences:

  	Beans	COM
  Events	Yes	Yes
  Properties	Yes	Yes
  Persistence	Yes	Yes
  Component Nesting	JDK 1.2	Yes, since OLE
  Drag-n-Drop	JDK 1.2	Yes, since OLE
  Introspection	Yes	Yes
  Tools	Few, many soon	Few, mostly MS
  Security Model	Tight (sandbox)	Loose (signing)

Reality Check

• Things are missing from Java and components, such as a consistent client platform

Recommendations

• The choice of distributed object protocols should be driven by the middleware support and client capabilities, not by protocol itself

The Future of Objects