2 - Distributed Information Systems Architecture

Question 1

Give an overview of distributed transactions

Answer 1

Keyword: COORDINATION

• Distributed is: subsystems working together in a COORDINATED fashion, possibly with multiple resource manager.
• Idea of distributed transactions: centralized COORDINATOR interacting with multiple agents (RM/DB).
• • Commit protocol: allow atomicity of global transactions (CID require extra protocols). Must provide:
• • • minimal overhead
• • • parallelism (independence) of agents
• • • fault tolerance (logging state changes and interactions).

Question 2

What are the goals of the Two-phase commit protocol and how does it work?

Answer 2

Perform commit on all agents in two phases:

1: receives EOT and send “prepare” message to all agents, waiting for a “failed” or “ready” response.
- Guarantees recoverability of changes

2: if all agents replied “ready”, send “commit” message to all of them, otherwise send “abort” message. Send ACK back to coordinator.
- Perform changes and release resources

• Every state change is logged on both coordinator and agent.
• Hierarchical/2PC for trees of requests

Question 3

What are the conceptual layers of an information system?

Answer 3

Presentation: accepts requests (input) and present information (output) (UI, reports, etc)

Application logic: implementation of services supported by the system (often retrieve/modify data)

Resource management: source of data. DB, files, “external” providers (encapsulation).

• Transactions span all layers.

Question 4

Compare the top-down and bottom-up approaches to the design of information systems.

Answer 4

TOP-DOWN
Ideal scenario, development from scratch
Steps of design/specification go from presentation down to necessary application logic (functionality) and data modelling.
Results usually in homogeneous tightly-coupled components.
High-level goals, addressing functional and non-functions requirements.

BOTTOM-UP
Necessity if reusing (RM), wrapping and integrating them on application logic, finally adapting presentation layer.
Functionality of components is predefined: high-level goals are achieved by adapting and combining them.
Results in loosely-coupled components, which remain autonomous and can be used stand-alone.

Question 5

How can the conceptual layers be mapped into tiers?

Answer 5

Tiers = implementation architecture

• Layers can be mapped to one or more tiers, as well as more layers into one tier (tier != node)
• Modularization of the system is achieved

Each tier provides well defined interfaces to access its functionality
Adding tiers improves flexibility, functionality, distribution, and scalability - but impacts performance and complexity.

Question 6

What are the different tier architectures and their pros and cons? 1-Tier

Answer 6

1-tier: monolith system

• Motivated by mainframe with scarce resources and dumb terminals: no other entry points (ie API)
• Legacy systems pre-middleware and DBMS
• Pros: optimized performance (tight coupling), closed environment
• Cons: difficult and expensive to maintain/reuse

Question 7

What are the goals and benefits of a horizontally distributed information system?

Answer 7

Reduced costs (vs. mainframe)
Organization separation, integration of existing components
Increase performance, scalability, reliability!

• Issues: distribution transparency to client, multiple realization alternatives.

Question 8

What are the communication alternative in the distributed information system scenario?

Answer 8

Blocking (synchronous) versus non-blocking (asynchronous) interactions

• Synch: caller must wait for reply before continuing processing (easier for programmer). Tighter coupling: receiver must be active.
• Asynch: caller continues processing: message queuing.

Question 9

What are the different tier architectures and their pros and cons? 2-tier

Answer 9

2-tier: client-server system

• motivated by powerful PCs/workstations coming up as clients
• Different approaches of where to place application logic (fat vs thin server and client): on client or on server
• Pros: emphasis on “service”: API (portability, stability by standardizing interface of server), supports multiple clients
• Cons: low scalability: central server, high-maintenance for fat clients (which also become integration engines for multiples servers)

Question 10

What are the different tier architectures and their pros and cons? 3-tier

Answer 10

3-tier: one tier for each IS layer (application logic introduces middleware)

• Pros: good scalability (application logic distributed across nodes), portability of application logic, supports multiple resource managers
• Cons: increased communication

Question 11

What are the different tier architectures and their pros and cons? N-tier

Answer 11

N-tiers: generalization of 3-tier (dividing singles layer into multiples tiers)

• Resource layer may include other tiered systems
• Presentation layer divided into client-side presentation and server-side presentation (4-tier architecture, presentation = browser + web server)

Question 12

Explain and compare the different alternative for distribution (for client transparency). ALTERNATIVE 1: unity of distribution = transaction

Answer 12

ALTERNATIVE 1: unity of distribution = transaction

• Route transaction to specific node (XOR), which are orthogonal on all tiers (sort of load-balancing)
• Pros: simple solution, easy to implement, heterogeneous environment (servers) supported
• Cons: inflexible, only local transactions (no distributed trans.)

Question 13

Explain and compare the different alternative for distribution (for client transparency). ALTERNATIVE 2: unit of distribution = application component

Answer 13

ALTERNATIVE 2: unit of distribution = application component (program) (class or library)

• Application components communicate among them (RPC/RMI), each one with its own local DB
• Allows distributed transactions
• Pros: locality of processing, resuse and heterogeneity
• Cons: inflexible data access (no DB operations across nodes), a bit more complex programming model (deals with distribution)

Question 14

Explain and compare the different alternative for distribution (for client transparency). ALTERNATIVE 3: unit of distribution = DB operation

Answer 14

ALTERNATIVE 3: unit of distribution = DB operation

• Application access local AND REMOTE DBs (multiple schemes/DBs, distributed transactions)
• Pros: high flexibility, operation across DBs (but not on single query) (see ALT4)
• Cons: complex prog. model (details of remote schemas needed), exposed heterogeneity, increased communication (for each query): lower performance
• • This ALT is worse than ALT2 because of performance/communication overhead

Question 15

Explain and compare the different alternative for distribution (for client transparency). ALTERNATIVE 4: distribution handled entirely by DB/RM

Answer 15

ALTERNATIVE 4: distribution handled entirely by DB/RM

• Single schema view for application (distributed/federated DBMS), distributed transactions
• Pros: query can join data from multiple sources, high flexibility for data access, simple and powerful programming model)
• Cons: increasing communication overhead, schema integration required.