Technical Stream Session 3: Design of Distributed Systems

1
Technical Stream Session 3Design of Distributed
Systems
Distributed Systems

• CSC 253
• Gordon Blair, François Taïani

2
Overview of the Session

• Applying the client-server model
• Application layering
• Two-tiered vs three-tiered architectures
• More general multi-tiered architectures
• Introducing threads
• Multi-threaded clients
• Multi-threaded servers
• Stateful vs. stateless servers
• Patterns for distributed systems programming

Associated Reading Tanenbaum and van Steen, pp.
42-53, 135-152
3
Application Levels

• Distributed applications commonly feature three
  levels
• The user-interface level
• Maintaining the interaction between the user and
  the application
• The processing level
• Often referred to as the business logic of the
  application
• The data level
• Managing the persistent state of the application
• Also responsible for consistency of data
• Often realised by a (generic) database supporting
  meta-data, queries, optimisation, etc

4
Two-Tiered Architectures

• Levels split between the a client and a server in
  various ways

5
Three-Tiered Architectures

• Same levels mapped on to three machines, for
  example

Application logic
(CF business logic)
Thin Client
Database Server
Application Server
6
2-Tiered vs 3-Tiered Architectures

• Advantages of 2-tiered approaches
• Advantages of three-tiered approaches

7
Multi-Tiered Approaches

• Take a given functional decomposition such as
  considered so far (user interface, processing,
  data)
• Vertical distribution
• Map the logical decomposition on to different
  machines
• Each machine then has its own distinct role
• Horizontal distribution
• Client or server functionality is physically
  split into logically equivalent units executing
  on different machines
• Very useful for load balancing and scalability
• Focus is then on coordination of co-operating
  elements

8
Introducing Threads

• What are threads?
• Lightweight units of concurrency operating in a
  single address space
• The multi-threaded client
• Latencies in a distributed system can be large,
  esp. in geographically large scale systems
• Threads can be used to hide such latencies
• A given thread blocks on an invocation
• One or more other threads can then proceed with
  computation or indeed other communication
  activity
• The multi-threaded server
• Assign threads to incoming invocations
• Supports the concurrent processing of requests
• Different thread-assignment policies possible
• Thread per request, tread per client, thread
  pool, etc
• Cf. object adapters

9
Example Policy
A multithreaded server organized in a
dispatcher/worker model
10
Stateless vs. Stateful Servers

• Stateless servers
• Do not keep state on clients and can change their
  own state without informing the client
• Web servers are the classic example of stateless
  servers
• Simplifies recovery after crashes
• Can lead to improved scalability (w.r.t. number
  of clients)
• Stateful servers
• Maintain state on each of their clients
• For example file servers are often (but not
  always) stateful in terms of maintaining
  information on clients (caching status,etc)
• Makes crash recovery more complex
• Can lead to performance optimisation

11
Patterns for Distributed Systems Programming

• The proxy pattern
• We have already seen the proxy pattern in the
  context of RPC/ RMI
• A proxy resides in the local address space and
  pretends to be the target (potentially remote)
  object
• In this way, the proxy can take steps to mask out
  location, access protocol, etc (i.e. implements
  basic distribution transparencies)
• Smart proxies
• A smart proxy extends this concept and allows
  arbitrary processing to be carried out, e.g.
  caching
• This effectively extends the range of
  transparencies that can be supported

12
Patterns (continued)

• Other key distributed systems patterns
• Interception
• Containment
• Other software patterns with applicability in
  distributed systems, for example
• Observer (cf. publish-subscribe)
• Façade (single object simplifying the interface
  to a set of objects)
• Mediator (provides a unified interface to a set
  of interfaces in a subsystem )
• Strategy (algorithms can be selected on-the-fly
  at runtime depending on conditions)
• Pipeline (cf. pipe and filter) consists of a
  chain of elements arranged so that the output of
  each element of the chain is the input of the
  next

See lectures on Aspects and EJB
13
Expected Learning Outcomes

• At the end of this third unit
• You should appreciate many of the subtleties and
  possibilities in distributed system design
• You should have at your disposal a plethora of
  techniques when approaching a distributed systems
  design
• You should understand the relative benefits of
  2-tiered, 3-tiered and multi-tiered architectures
• You should understand the potential of
  multi-threaded architectures in the
  implementation of clients and servers
• You should be able to compare and contrast
  stateless vs. stateful approaches to server
  design
• You should be aware of a range of patterns as
  applicable to distributed system design