高级软件工程 - SJTUwang-xb/wireless_new/coursePages/...Google GFS; Hadoop HDFS; Kosmix KFS...
Transcript of 高级软件工程 - SJTUwang-xb/wireless_new/coursePages/...Google GFS; Hadoop HDFS; Kosmix KFS...
高级软件工程
第9节课:云计算和大数据
主讲:刘驰
2013年11月18日
Content
Cloud ComputingHadoopHDFSMapReduceHBase
3
云计算的相关概念云计算的定义
云计算的特征
云计算的分类
云计算与其它计算比较
云计算的优势与带来的变革
4
为什么需要云计算?
案例一:
写文件
电脑硬盘坏了,文件丢失
存储在云中的文件不会丢失
5
为什么需要云计算?
案例二:
QQ聊天 ‐‐‐下载、安装、使用
使用C++‐‐‐下载、安装、使用
……
从云中获取服务
6
为什么需要云计算?
案例三:
华盛顿邮报突然需要大量计算资源进行文件格式转化
报社现有计算能力每页需要30分钟
新闻时效性不允许
使用Amazon EC2的计算资源
77
云计算产生的原动力
8
为什么叫云计算?
云‐‐‐互联网?透明性?云里雾中,不得其解?
9
什么是云计算?
10
什么是云计算?
11
云计算系统构架
12
云计算的特征
13
云计算的分类
按服务类型
14
云计算分类
服务类型分类
分类 服务类型 运用灵活性 运用难易程度
IaaS 接近原始的计算存储能力 高 难
PaaS 应用的托管环境 中 中
SaaS 特定功能 低 易
15
云计算分类
不同类型云的案例
16
云计算分类
按服务方式分
17
云计算的相关概念
云计算与其它计算比较云计算与并行计算
云计算与网格计算
云计算与效用计算
云计算的优势与带来的变革
18
云计算与并行计算
并行计算(高性能计算、超级计算)一群同构处理单元的集合,这些处理单元通过通信和协作来更快地解决大规模计算问题
19
20
云计算与网格计算
网格计算(分布式计算)将分散在网络中的空闲服务器、存储系统和网络连接起来,形成一个整合系统,为用户提供功能强大的计算及存储能力来处理特定的任务
21
22
云计算与效用计算
Utility Computing: IT资源能够根据用户的要求按需
提供,并根据使用情况付费
23
云计算与其它类型计算
24
云计算的相关概念
云计算与其它计算比较
云计算的优势与带来的变革云计算的优势
云计算带来的变革
云计算产生的原动力
25
云计算的优势
优化产业布局例:Google数据中心分布
26
云计算的优势
推进专业分工例:中小公司的数据中心 vs专业公司的大型数据中心
数据中心属性 中小型数据中心 大型数据中心
服务器个数 <2000 >20000
每个管理员管理服务器数
<500 >500
PUE值 2.0~2.5 1.0~1.5
服务器供电方式 交流电 直流电
电价 高 低
制冷方式 风冷 水冷+风冷
提供单位计算力的成本 高 低27
云计算的优势
提升资源利用力例:新兴公司将IT业务外包给专业的云计算提供商提供
管理
28
云计算的优势
减少初期投资例:企业使用云中的计算资源和服务,无需购买硬件和授权
29
云计算的优势
降低管理开销应用管理的动态、高效率、自动化
30
云计算带来的变革
机遇与挑战云计算产业结构中的角色
31
云构架层次
共有云(通过Internet提供公共服务)
混合云(通过Internet和Intranet提供公共
和私有服务)
私有云(通过Intranet提供私有服务)
应用层软件及服务(SaaS)
平台层平台即服务(PaaS)
基础设施层基础设施即服务(Iaas)
32
云构架的服务层次
基础设施即服务提供虚拟化的计算资源、存储资源、网络资源Amazon EC2
平台即服务使开发人员充分利用开放资源来开发定制应用Google AppEngine
软件即服务软件或应用通过租用的形式提供给用户使用Salesforce.comGoogle Gmail、Docs
33
IaaS基本功能
34
Example: Amazon EC2
底层采用Xen虚拟化技术,以Xen虚拟机的形式向
用户动态提供计算资源
按照用户使用资源的数量和时间计费
http://aws.amazon.com/ec2/
35
PaaS开发测试环境
应用模型、API代码库、开发测试环境
运行时环境验证、配置、部署、激活
运营环境升级、监控、淘汰、计费
36
Example: Google App Engine在Google的基础架构上运行自己的网络应用程序
提供网址抓取、邮件、memcache、图像操作、计
划任务等服务
目前支持Java和Python
3737
SaaS可以通过浏览器访问,具有开放的API
在使用的过程中根据实际使用情况付费
较强的云应用之间的整合能力
38
SaaS分类
标准应用
如文档处理、电子邮件、日程管理等提供商往往是实力雄厚的IT业巨头
客户应用如客户管理系统CRM、企业资源计划系统ERP提供商是规模较小的专业公司
多元应用如地铁时刻表服务Mutiny、期权交易方案提供The Option Lab提供商多是规模较小的开发团队
39
Example: Google Docs & Docs for Facebook
• 在线文档编辑• 多人协作编辑
40
Web QQ一站式网络服务
41
Salesforce.com客户应用的典型代表
采用了多租户的架构
所有用户和用户和应用程序共享一个实例,同时又能够按需满足不同的客户要求
42
An Ecosystem for Cloud Computing
43
Problem
Batch (offline) processing of huge data set using commodity hardware is not enough for real‐time applications
Strong desire for linear scalability
Need infrastructure to handle all mechanics
allow developers to focus on the processing logic/algorithms
44
Explosive Data! – StorageNew York Stock Exchange: 1 TB data per day
Facebook: 100 billion photos, 1 PB (1000 TB)
Internet Archive: 2 PB data, growing by 20 TB per month
Can’t put data on a SINGLE node
Strong needs for distributed file systems
46Java/Python/C interfaces
47
Commercial Hardware
典型的2层构架–节点是普通的商业PC机– 30‐40 节点/Rack–顶层到Rack 带宽3‐4Gbps– Rack到节点带宽1Gbps 48
Hadoop HistoryDec 2004 – Google GFS 论文发表
July 2005 – Nutch使用MapReduceFeb 2006 –成为 Lucene子项目
Apr 2007 – Yahoo! 建立 1000个节点的集群
Jan 2008 –成为 Apache顶级项目
Jul 2008 –建立 4000 节点的测试集群
Sept 2008 – Hive 成为Hadoop子项目
……49
Who is Using Hadoop?
50
Example: Facebook的Hadoop集群
产品集群
4800个内核,600个机器,每个机器16GB—2009年4月8000个内核,1000个机器,每个机器32GB—2009年7月每个机器拥有4个1TB大小的SATA硬盘
两层网络结构,每个Rack有40个机器
整个集群大小为2PB,未来还会不断增加
测试集群
• 800 个内核, 每个16GB
51
A Distributed File System
Single‐Node Architecture
Memory
Disk
CPU
Machine Learning, Statistics
“Classical” Data Mining
53
Commodity ClustersWeb data sets can be very large
Tens to hundreds of TBCannot mine on a single serverStandard architecture emerging:
Cluster of commodity Linux nodesGigabit Ethernet interconnect
How to organize computations on this architecture?
Mask issues such as hardware failure 54
Cluster Architecture
Mem
Disk
CPU
Mem
Disk
CPU
…
Switch
Each rack contains 16‐64 nodes
Mem
Disk
CPU
Mem
Disk
CPU
…
Switch
Switch1 Gbps between any pair of nodesin a rack
2‐10 Gbps backbone between racks
55
Stable storageFirst order problem: if nodes can fail, how can we store data persistently? Answer: Distributed File System
Provides global file namespaceGoogle GFS; Hadoop HDFS; Kosmix KFS
Typical usage patternHuge files (100s of GB to TB)Data is rarely updated in placeReads and appends are common
56
57
58
Namenode and DatanodesMaster/slave architecture1 Namenode, a master server that manages the file system namespace and regulates access to files by clients.many DataNodes usually one per node in a cluster.
manage storageserves read, write requests, performs block creation, deletion, and replication upon instruction from Namenode.
HDFS exposes a file system namespace and allows user data to be stored in files.A file is split into one or more blocks and set of blocks are stored in DataNodes.
2013/11/1859
Namespace
2013/11/18
Hierarchical file system with directories and filesCreate, remove, move, rename etc.Namenode maintains the file systemAny meta information changes to the file system recorded by the Namenode.An application can specify the number of replicas of the file needed: replication factor of the file. This information is stored in the Namenode.
60
Data Replication
2013/11/18
Store very large files across machines in a large cluster.Each file is a sequence of blocks of same size.Blocks are replicated 2‐3 times.Block size and replicas are configurable per file.Namenode receives a Heartbeat and a BlockReport from each DataNode in the cluster.BlockReport contains all the blocks on a Datanode.
61
Replica Placement
2013/11/18
Rack‐aware: Goal: improve reliability, availability and network bandwidth utilizationResearch topic
Namenode determines the rack id for each DataNode.Replicas are placed: 1 in a local rack, 1 on a different node in the local rack and 1 on a node in a different rack.1/3 of the replica on a node, 2/3 on a rack and 1/3 distributed evenly across remaining racks.
62
HDFS: Data Node Distance
63
Replication PipeliningWhen the client receives response from Namenode, it flushes its block in small pieces (4K) to the first replica, that in turn copies it to the next replica and so on.Thus data is pipelined from Datanode to the next.
2013/11/1864
Replica Selection
2013/11/18
Replica selection for READ operation: HDFS tries to minimize the bandwidth consumption and latency.If there is a replica on the Reader node then that is preferred.HDFS cluster may span multiple data centers: replica in the local data center is preferred over the remote one.
65
Datanode
2013/11/18
A Datanode stores data in files in its local file system.Datanode has no knowledge about HDFS filesystemIt stores each block of HDFS data in a separate file.Datanode does not create all files in the same directory.It uses heuristics to determine optimal number of files per directory and creates directories appropriately:
Research issue?
When the filesystem starts up it generates a list of all HDFS blocks and send this report to Namenode: Blockreport.
66
HDFS: File Read
67
HDFS: File Write
68
Communication Protocol
2013/11/18
All protocols are layered on top of the TCP/IP protocolA client establishes a connection to a configurable TCP port on the Namenode machine. It talks ClientProtocol with the Namenode.Datanodes talk to the Namenode using Datanode protocol.RPC abstraction wraps both ClientProtocol and Datanodeprotocol.Namenode is simply a server and never initiates a request; it only responds to RPC requests issued by DataNodes or clients.
69
DataNode Failure and HeartbeatDatanodes lose connectivity with Namenode.Namenode detects this condition by the absence of a Heartbeat message.Namenode marks Datanodes without Hearbeat and does not send any IO requests to them.Any data registered to the failed Datanode is not available to the HDFS.
2013/11/1870
Cluster RebalancingHDFS architecture is compatible with data rebalancing schemes.A scheme might move data from one Datanode to another if the free space on a Datanode falls below a certain threshold.In the event of a sudden high demand for a particular file, a scheme might dynamically create additional replicas and rebalance other data in the cluster.These types of data rebalancing are not yet implemented: research issue.
2013/11/1871
APIsHDFS provides Java API for application to use.Python access is also used in many applications.A C language wrapper for Java API is also available.A HTTP browser can be used to browse the files of a HDFS instance.
2013/11/1872
FS Shell, Admin and Browser InterfaceHDFS organizes its data in files and directories.It provides a command line interface called the FS shell that lets the user interact with data in the HDFS.The syntax of the commands is similar to bash and csh.Example: to create a directory /foodir
/bin/hadoop dfs –mkdir /foodirThere is also DFSAdmin interface availableBrowser interface is also available to view the namespace.
2013/11/1873
A Distributed Computation Framework for Large Data Set
What is Map/Reduce?A Programming Model
Decompose a processing job into Map and Reducestages
Developer need to provide code for Map and Reduce functionsconfigure the joblet Hadoop handle the rest
75
MapReduce Model
76
Architecture Overview
Job tracker
Task tracker Task tracker Task tracker
Master Node
Slave node 1 Slave node 2 Slave node N
Workers
user
Workers Workers77
Inside Hadoop
78
Warm up: Word CountWe have a large file of words, one word to a lineCount the number of appearances for each distinct word
Sample application: analyze web server logs to find popular URLs
79
Word Count (2)Case 1: Entire file fits in memoryCase 2: File too large for mem, but all <word, count> pairs fit in memCase 3: File on disk, too many distinct words to fit in memory
sort datafile | uniq –c
80
Word Count (3)To make it slightly harder, suppose we have a large corpus of documentsCount the number of times each distinct word occurs in the corpuswords(docs/*) | sort | uniq -cwhere words takes a file and outputs the words in it, one to
a line
The above captures the essence of MapReduceGreat thing is it is naturally parallelizable
81
MapReduceInput: a set of key/value pairsUser supplies two functions:
map(k,v) list(k1,v1)reduce(k1, list(v1)) v2
(k1,v1) is an intermediate key/value pairOutput is the set of (k1,v2) pairs
82
What is MAP?Map each data entry into a pair <key, value>
ExamplesMap each log file entry into <URL,1>Map day stock trading record into <STOCK, Price>
83
What is Shuffle/Merge phase?Hadoop merges(shuffles) output of the MAP stage into<key, valulue1, value2, value3>
Examples<URL, 1 ,1 ,1 ,1 ,1 1><STOCK, Price On day 1, Price On day 2..>
84
What is Reduce?
Reduce entries produces by Hadoop merging processing into <key, value> pair
ExamplesMap <URL, 1,1,1> into <URL, 3>Map <Stock, 3,2,10> into <Stock, 10>
85
Pseudo‐Code: Word Countmap(key, value):// key: document name; value: text of document
for each word w in value:emit(w, 1)
reduce(key, values):// key: a word; values: an iterator over counts
result = 0for each count v in values:
result += vemit(key,result) 86
map(key=url, val=contents):For each word w in contents, emit (w, “1”)reduce(key=word, values=uniq_counts):
Sum all “1”s in values listEmit result “(word, sum)”
see bob runsee spot throw
see 1bob 1 run 1see 1spot 1throw 1
bob 1 run 1see 2spot 1throw 1
87
Word Count
Example uses: distributed grep distributed sort web link-graph reversal term-vector / host web access log stats inverted index construction
document clustering machine learning statistical machine translation
... ... ...
Widely ApplicableMapReduce Programs in Google Source Tree
88
• 100s/1000s of 2‐CPU x86 machines, 2‐4 GB of memory • Limited bisection bandwidth • Storage is on local IDE disks • GFS: distributed file system manages data (SOSP'03) • Job scheduling system: jobs made up of tasks, scheduler assigns tasks to machines
Implementation is a C++ library linked into user programs
Implementation Overview
89
Distributed Execution Overview User
Program
Worker
Worker
Master
Worker
Worker
Worker
fork fork fork
assignmap
assignreduce
readlocalwrite
remoteread,sort
OutputFile 0
OutputFile 1
write
Split 0Split 1Split 2
Input Data
90
Data FlowInput, final output are stored on HDFS
Scheduler tries to schedule map tasks “close”to physical storage location of input data
Intermediate results are stored on local FS of map and reduce workersOutput is often input to another map reduce task
91
CoordinationMaster data structures
Task status: (idle, in‐progress, completed)Idle tasks get scheduled as workers become availableWhen a map task completes, it sends the master the location and sizes of its R intermediate files, one for each reducerMaster pushes this info to reducers
Master pings workers periodically to detect failures 92
FailuresMap worker failure
Map tasks completed or in‐progress at worker are reset to idleReduce workers are notified when task is rescheduled on another worker
Reduce worker failureOnly in‐progress tasks are reset to idle
Master failureMapReduce task is aborted and client is notified
93
Execution
94
Parallel Execution
95
How Many Map and Reduce Jobs?M map tasks, R reduce tasksRule of thumb:
M, R >> (# of nodes) in clusterOne DFS chunk per map is commonImproves dynamic load balancing and speeds recovery from worker failure
Usually R is smaller than M, because output is spread across R files
96
CombinersOften a map task will produce many pairs of the form (k,v1), (k,v2), … for the same key k
e.g., popular words in Word CountCan save network time by pre‐aggregating at mapper
combine(k1, list(v1)) v2same as reduce function
97
Partition FunctionInputs to map tasks are created by contiguoussplits of input fileFor reduce, we need to ensure that records with the same intermediate key end up at the same workerSystem can use a default partition function e.g., hash(key) mod RSometimes useful to override
e.g., hash(hostname(URL)) mod R ensures URLs from a host end up in the same output file 98
Execution SummaryHow is this distributed?
1. Partition input key/value pairs into chunks, run map() tasks in parallel
2. After all map()s are complete, consolidate all emitted values for each unique emitted key
3. Now partition space of output map keys, and run reduce() in parallel
If map() or reduce() fails, re‐execute!
99
Example: Trading Data Processing
Input: Historical Stock DataRecords are CSV (comma separated values) text file Each line : stock_symbol, low_price, high_price1987‐2009 data for all stocks one record per stock per day
Output:Maximum interday delta for each stock
100
Map Function: Part I
101
Map Function: Part II
102
Reduce Function
103
Running the Job : Part I
104
Running the Job: Part II
105
A Distributed Storage System
What is HBase?
Distributed Column‐Oriented database on top of HDFS
Modeled after Google’s BigTable data store
Random Reads/Writes on sequential stream‐oriented HDFS
Billions of Rows * Millions of Columns * Thousands of Versions
107
Where is HBase?
HBase is built on top of HDFS
HBase files are internally
stored in HDFS
108
Logical View
Row Key Time Stamp
Column Contents
Column Family Anchor (Referred by/to)
Column “mime”
“com.cnn.www”
T9 cnnsi.com cnn.com/1
T8 my.look.ca cnn.com/2T6 “<html>.. “ Text/htmlT5 “<html>.. “t3 “<html>.. “
109
Physical View
Row Key Time Stamp Column: ContentsCom.cnn.www T6 “<html>..”
T5 “<html>..”T3 “<html>..”
Row Key Time Stamp Column Family: AnchorCom.cnn.www T9 cnnsi.com cnn.com/1
T5 my.look.ca cnn.com/2
Row Key Time Stamp Column: mimeCom.cnn.www T6 text/html
110
Region Servers
Tables are split into horizontal regionsEach region comprises a subset of rows
HDFSNamenode, dataNode
MapReduceJobTracker, TaskTracker
HBASEMaster Server, Region Server
111
HBASE Architecture
112
HBase vs. RDMS
HBase tables are similar to RDBS tablesDifferences:Rows are sorted with a Row KeyColumns can be added on‐the‐fly by client as long as the column family they belong to pre‐exists
113