Practical exercise unit 7 Topic : MapReduce program in Java Work purpose : Display function, convolution function and the assignment perform the code study

Nodes name and node information
The HDFS cluster is in a master-slave scheme working two kind of from nodes consists of : Nodes name ( manager ) and one how many data nodes ( subordinates ). Name nodes file system namespace _ manages _ It is a file system the tree and in the tree all files and of directories metadata keeps _ Information local on disk two file as stored : namespace image and changes magazine . Name nodes which data nodes given of the file all blocks he knows how to keep it ; but , of blocks location about data constant respectively not saved , but the system to work on unloading data of nodes information according to again will be restored .
User on behalf file to the system access for customer name and data nodes with is connected . The client is similar to POSIX file system interface present ( Portable Operating System Interface ), therefore for user code name nodes and data nodes about never thing to know a must not _
Node data - file system main " worker power " is. They blocks they read and write ( customers at your request according to or name nodes ) and from time to time with themselves which keeps blocks list to the name node they transmit
Nod is a nodule file system not working remains . From this except , if on the name node working car no done if , the file in the system all files disappears because _ them data nodes from the blocks recovery possible it won't be . Therefore , the name node possible has been to failures resistant to do important ; This is Hadoop the problem solution of doing two mechanism present is enough
First method is file system of metadata stable situation defines files archiving . of the Hadoop namenode constant situation one how many file to systems to write for setup can _ Writing deeds synchronous and atomic way will be done . Usually , the configuration local to the driver and to a remote NFS volume to write for made up .
Secondary name node to work it is also possible to download , it is its own in the name of despite the name node task doesn't do it . His main task occasionally namespace _ _ _ image changes to the magazine combine , his too much except increase prevention is to get Secondary name nodes usually separately physical in the car because it works combine important again work power and to the name node equal to requires memory . _ Secondary name nodes combined names area of the image a copy keeps it _ main node not working while remaining use can _
HDFS Federation
the name node file system all files and to the blocks links in memory stored ; this a lot to files have has been very big in clusters memory of scaling restrictive to the factor rotation means ( See "Name node " on page 398 how much memory (see " need ?" ).
HDFS technology 2.x releases in the series present done Federation name nodes to add through the cluster scaling opportunity gives _ their each one file of the system namespace one part manages _ For example , one named node / user in the hierarchy all files , second named node while / share exchange in the hierarchy files manage can _
Empty in place all file blocks own into received blocks from the pool consists of namespace _ size manages _ Names area volumes from each other independent ; these are name nodes each other with contact that he won't means and of a single name node not working stay another nodes by of managed namespaces to existence effect does not However , blocks pool is not divisible , therefore for data nodes in the cluster each in a name node register passes and different different in pools blocks keeps _
HDFS federation to the cluster access for customers nodes naming - for file ways mapping for installation from tables they use This task ViewFileSystem using configuration level solution will be done and ViewFileSystem i URI viewfs ://.