lec1
building and classifying parallel and distributed systems
bus-based multiprocessors
SMP:symmetric multi-processing
characterized by UMA:uniform memory access
cache: snoopy
not scalable
switched multiprocessors
crosspoint switches–expensive switching fabric
numa
non uniform memory access
distributed systems
- no shared clock
no shared memory(use network interconnect)
bus-based multicomputers
- switched multicomputers(any to any communication)
distribured software service models
multiprocessor/multicore os
network os
distribution is explicit
distributed os
奇葩
middleware
why building distributed systems
scalability
consideations:
- components(users,data,cpus):overloaded
- geography:communication latency,bandwidth,reliability
- administration:complexity
design issues
- reliability
- availability: fraction of time system is usable
- reliability: data must not get lost
- performance
- communication network may be slow and/or unreliable
scalability
- distributable vs centralized alg
service model
centralized model
limiting factor: cpus in system
client-server model
service client server
peer-to-peer model
equivalent capabilities
no machines dedicated to serving others
processor pool model
collection of cpus that can be assigned processes on demand
cloud computing model
software as a service
platform as a service:平台作为服务,由用户选择操作系统和中间软件,供应商提供硬件软件资源,来控制资源的使用
infrastructure as a service:基础设施作为服务,如何控制资源也交给用户