Distributed System. Gang Wu. Spring,2019

Transcription

1 Distributed System Gang Wu Spring,2019

2 Computer Systems How a single computer works? (traditional single core system) Interrupt User-level System-level Operating System How a distributed system works? (Many computers work together) How to communicate How to store and manage data How to compute How to cooperate

3 What is Distributed System Flynn s Taxonomy on Computer Architecture (1966) SISD: Traditional uniprocessor system SIMD: Array (vector) processor, for Example GPU, APU, SSE3, 3D NOW! MISD: Generally not used and doesn t make sense MIMD: Parallel and Distributed systems Single Data Single Instruction SISD Multiple Instruction MISD Multiple Data SIMD MIMD

4 What is Distributed System classifying MIMD Shared memory systems --- multiprocessors (UMA...) No shared memory --- multicomputers (MPP, COW...) SMP Server Cluster

5 Why Distributed is important Moore's Law the empirical observation that transistor density in a microprocessor doubles every 18 to 24 months power consumption P = C V 2 F additional transistors for parallel computing (MultiCore) Your computer has become a parallel/distributed system

6 Why Distributed is important In the Internet world Most machines are distributed and connected Web Application icloud WeChat Internet Game...

7 Distributed system A distributed system consists of multiple autonomous computers that communicate through a computer network. The computers interact with each other in order to achieve a common goal. There are several autonomous computational entities, each of which has its own local memory The entities communicate with each other by imformation passing

8 Distributed Software Service Models How to deal with distributed? Multiprocessor/Multicore OS Network OS Distributed OS Middleware

9 Distributed Software Service Models Multiprocessor/Multicore OS Multiple CPUs: UMA/NUMA #Processors transparent to programs Single OS and system image Same communication mechanisms Semaphores, shared memory, messages Networking not needed We have shared memory Operating System CPU CPU CPU

10 Distributed Software Service Models Network OS Autonomous nodes comm. via a network Configurations may differ Not a single system image Distribution is explicit OS OS OS Examples: Linux, Windows, OS X CPU CPU CPU Network

11 Distributed Software Service Models Distributed OS Operating systems are aware of multiple systems and coordinate resource sharing Single system image for file systems, processes, devices Homogeneous hardware Distributed OS CPU CPU CPU Network

12 Distributed Software Service Models Middleware User-level software (libraries): layer on top of network OS Provides common interface for transparency of specific functions Often OS-agnostic Examples: PRC, RMI,.NET, MapReduce OS CPU Middleware OS CPU OS CPU Network

13 Goals of the cource Introduce you the distributed system design concepts and principles Computing model, Layering, Reliability, Consistency,... Learn and practice many important popular distributed systems Hadoop (Batch processing) Spark (Stream processing) Learn and practice how to deal with a tough real problem Such as SecKill in an online shopping

14 Topics in distributed systems File Systems Programming Model Storage Transaction

15 Discuss some topics Reliability System Availability: Reliability & Maintainability Data Durability: data must not get lost Performance Communication network may be slow and/or unreliable TCP vs UDP Scalability Distributable vs. Centralized algorithms Can we take advantage of having lots of computers?

16 Topic we must talk Failure is the defining difference between distributed and local programming, so you have to design distributed systems with the expectation of failure "Failure happens all the time. It is your number one concern. Hardware (before the late 80's) and software failure (25%-35% even with rigorous testing) You know you have a distributed system when the crash of a computer you've never heard of stops you from getting any work done. -- Leslie Lamport

17 Course schedule Design concepts and principles Intro DS RPC & Messaging Consistency Failure Processing Discuss:What will happen when DS meet Large Scale? Hadoop & Spark Map/Reduce, Yarn, HDFS, Hbase, ZooKeeper Spark, Spark Straming, Spark Mlib Some other useful systems memcached, Kafka

18 Computing service models Centralized Model No networking Traditional time-sharing system Direct connection of user terminals to system One or several CPUs Not easily scalable Limiting factor: #CPUs in system Contention for same resource

19 Computing service models Client-server Model Environment consists of clients and servers Service: task machine can perform Server: machine that performs the task Client: machine that is requesting the service 2-tier/3-tier/multi-tier Client-server model

20 Computing service models Peer-to-peer Model Each machine on network has (mostly) equivalent capabilities No machines are dedicated to serving others Decentralization Structed P2P(DHT),unstructured P2P(Random) E.g. collection of PCs: Send/receive (without server) Peer-to-peer file sharing PPLive computation

21 Computing service models Cloud Computing Model Provide users with seamless access to Computation/Storage/Communication Heterogeneous and geographically distributed systems Build a supercomputer on the fly via networked, loosely coupled computers Offerings exist at various levels of abstraction Infrastructure as a Service (IaaS) e.g., Amazon EC2, AliCloud Platform as a Service (PaaS) e.g., Windows Azure, Google AppEngine Software as a Service (SaaS) e.g., Salesforce.com

22 Typical Datacenter Applications Single big computation (e.g. big data) Hosted apps for many tenants (e.g. Gmail, web hosting) Single big multi-user app (e.g. Facebook)