T - Aula 2

🌫 18 Setembro 2023 - #ASCN

Why distributed systems?

• Modularity, decoupling different concerns.
• Performance.
• Dependability

How to distribute?

1. Monolythic system

• Architecture: Monolithic system.excalidraw
• Multiple services for multiple targets in the same server

2. Distributed systems

• Main distribution concerns:
  1. Replication
  2. Partitioning
  3. Service-orientation
• All of these address scaling out a service/application.
• Not mutually exclusive, can be combined.

2.1 Replication

• Architecture: Replication.excalidraw
• Multiple copies of the same data and functionality
• Addresses resilience and scale-out

2.2 Partitioning

• Architecture: Partitioning.excalidraw
• A server is split horizontally (Sharding).
• Addresses scale-out.
• Can be applied to computation, data, ...

2.3 SOA - Service-Oriented Architecture

• Architecture: SOA.excalidraw
• Addresses scale-out and modularity.
• Example: micro-services.

2.3.1 Microservices

• Each service implements specific functionality.
• Services can scale independently.
• Decomposition may be troublesome: how micro is micro?
• Consistency.
• Complex deployment and testing

Distributed architectures

1. Client-server

• Architecture: client-server.excalidraw
• Functionality and data are in the server.
• A stub runs embedded in the client.
• The stub is part of the server software package
  • E.g., the Web (“protocol” is HTTP)

2. Proxy Server

• Architecture: proxy.excalidraw
• Multiple servers can be used transparently.
• The proxy is a performance and availability bottleneck.
  • E.g. MongoDB

3. Master Server

• Architecture: master.excalidraw
• Scale out!
  • E.g. HDFS

4. Server Group

• Architecture: server-group.excalidraw
• All servers can process requests.
• Coordination may be necessary.
• Resiliency!
  • E.g. ZooKeeper

5. Bus

• Architecture: bus.excalidraw
• The bus routes messages.
• Participants publish and consume messages to/from the bus.
• Decouples producers from consumers.
• Flexibility!
  • E.g. Kafka

6. Multi-tier

• Architecture: multi-tier.excalidraw
• Each server acts as a client of the next tier.
• Allows independent deployment and scaling of different functionality.
• E.g. AS + DBMS:
  • “protocol A” == Web (e.g.)
  • “Stub B” == Database Driver!
  • “protocol B” uses SQL

6.1 State in multi-tier

• Persistent state is harder to replicate and shard.
• Computation is easier to replicate and shard.
• No state in upper tiers:
  • Web browser
• Transient / cached state in middle tiers:
  • Application server
• Persistent state at lower tiers:
  • Database