Lecture 02
2023-11-21
- Section 01 (Cont.)
- System cycles
- Bandwidth vs. computation
- High bandwidth => Delegate job to another machine
- Cheap computation => thick client
- Grid computing:
- For improving throughput
- Grid system architecture, standards and middleware
- Overlay network (logical level) & IP network (Physical level)
- Distribute system does not share memory
- Cloud & IoT
- IoT: A thing with processor and network hardware
- Three layers: Cloud, Edge and Sensors (Controllers)
- Fog: layer between cloud and edge
- Numbers of devices: Edge => Billions, Fog => millions, cloud => thousands
- Distributed parallel computing system classification
- Multicomputer cluster: Big box of computers
- Pear-to-Pear: Not for high performance computing (PC's are too far away), easy to scale, but hard to manage, hard to know state of the whole system.
- Client-Server: Server often becomes a bottleneck (does not scale), but easy to manage. Easy to take a picture of and coordinate system state.
- Share memory vs. distributed memory
- Processes operate on data, exchange data with each others
- Two ways of communication
- Shared memory: One process place data on some place, another process reads it. Memory is centralized and not scalable. Variable "A" here and "A" there are the same.
- Distributed memory: Variable "A" here and "A" there are not the same. Communicate through network.
- Splitting processes and memories introduces new complexity dimension
- Splitting processes: HPC
- Splitting memories: Big Data
- Parallel and distributed programming
- MPI: Message passing interface (focus on processes)
- We are at a state with a lot of data (data querying)
- MapReduce: a programing paradigm (not a software)
- MPI: Message passing interface (focus on processes)
- Distributed System: Definitions
- Distributed vs. Paralleled
- Not necessarily depends on physical layer (split processes / split memories / computer network)
- Definition: Network + only communicate with message passing
- Network layer review
- Computer network vs. Distributed System
- Data Transmission | Computer network infrastructure | Distributed System
- Distributed System challenges
- Transparencies: make some concepts invisible to a user
- System cycles