What is Spark?

1,Apache Spark is a fast and general engine for large-scale data processing

2,Speed

3,Ease of Use

4,Generality

5,Integrated with Hadoop

 

One Stack to rule them all:        Spark Streaming     ======>     Ad-hoc Queries

                          \\      SPARK      //

                           \\              //

                            =>  Batch Processing <=

 

Hadoop Data Sharing And Spark Data Sharing:

ps:Hadoop频繁与磁盘上的数据进行交互,频繁对数据进行序列化和反序列化

 

Hadoop的Map Reduce类比图

 

Why Spark fast ?  

1,Memory based computation

2,DAG   有向无环图(Directed Acyclic Graph)

3,Thread model

4,Optimization (eg:delay scheduling)

 

API: Scala  Python  Java  R

 

Cluster Manager: Local  Standalone  Yarn  Mesos

 

 

Dependency:norrow  wide