Scribd
Upload
11 views57 pages

Chapter 6

Chapter 6 of 'Distributed and Cloud Computing' discusses Cloud Programming and Software Environments, focusing on the MapReduce model for large-scale data processing. It explains how MapReduce simplifies data processing across distributed systems by utilizing user-defined functions for mapping and reducing data, and highlights its applications in various fields such as search indexing and data mining. The chapter also covers the architecture of MapReduce, fault tolerance mechanisms, and the Hadoop framework as an implementation of this model.

Uploaded by

bhanumathis
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
11 views57 pages

Chapter 6

Chapter 6 of 'Distributed and Cloud Computing' discusses Cloud Programming and Software Environments, focusing on the MapReduce model for large-scale data processing. It explains how MapReduce simplifies data processing across distributed systems by utilizing user-defined functions for mapping and reducing data, and highlights its applications in various fields such as search indexing and data mining. The chapter also covers the architecture of MapReduce, fault tolerance mechanisms, and the Hadoop framework as an implementation of this model.

Uploaded by

bhanumathis
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
You are on page 1/ 57
Distributed and Cloud Computing K. Hwang, G. Fox and J. Dongarra Chapter 6: Cloud Programming and Software Environments ela a Adapted from Kai Hwang, University of Southern California TTT lel tim UCD} Matei Zaharia, EECS, UC Berkeley November 25, 2012 Cr Uae ad ry asa Parallel Computing and Programming Enviroments e MapReduce e Hadoop e Amazon Web Services What is MapReduce? e Simple data-parallel programming model e For large-scale data processing > Exploits large set of commodity computers > Executes process in distributed manner pm Ohi -1e-M ale) a= NZTICLe)I LANs e Pioneered by Google > Processes 20 petabytes of data per day e Popularized by open-source Hadoop project » Used at Yahoo!, Facebook, Amazon, ... What is MapReduce used for? e At Google: » Index construction for Google Search > Article clustering for Google News » Statistical machine translation e At Yahoo!: > “Web map” powering Yahoo! Search > Spam detection for Yahoo! Mail e At Facebook: > Data mining > Ad optimization >» Spam detection Motivation: Large Scale Data Processing e Many tasks composed of processing lots of data to produce lots of other data e Want to use hundreds or thousands of CPUs .. but this needs to be easy! e MapReduce provides > va v User-defined functions Automatic parallelization and distribution Lar Oise) (ie lal) 1/0 scheduling Status and monitoring What is MapReduce used for? e Inresearch: Astronomical image analysis (Washington) Bioinformatics (Maryland) Analyzing Wikipedia conflicts (PARC) Natural language processing (CMU) Particle physics (Nebraska) Ocean climate simulation (Washington) vVvVvVVVYV Vv Na Distributed Grep — grep —- Very [Spit data ee matches | em Mesiaecew —- grep —- ad] in athe | data . . Syigecicy — grep —~ grep is a command-line utility for searching plain-text data sets for lines matching a regular expression. cat is a standard Unix utility that concatenates and lists files Distributed Word Count Split data Feemuellioee| count | Very Split data eee count | "4 merge big [geal Spit data PEC count MMC co cnr data . 2 Fy Fy EEIEEE —- count— Very big data Partitioning Function IV ETeR Accepts input key/value pair Emits intermediate key/value pair Reduce : Accepts intermediate key/value* pair Emits output key/value pair 9 Architecture overview Master node ig ESET alele AN Slave node SEV allel 4 GFS: underlying storage system e Goal > global view >» make huge files available in the face of node failures e Master Node (meta server) > Centralized, index all chunks on data servers e Chunk server (data server) > File is split into contiguous chunks, typically 16- Ee > Each chunk replicated (usually 2x or 3x). Try to keep replicas in different racks. GFS architecture Chunkserver 1 Chunkserver 2 Chunkserver N Functions in the Model Map >» Process a key/value pair to generate intermediate key/value pairs Reduce > Merge all intermediate values associated with the Tals Partition > Bydefault: hash(key) mod R > Well balanced Programming Concept Map > Perform a function on individual values in a data set to create a new list of values > Example: square x = x * x map square [1,2,3,4,5] returns [1,4,9, 16,25] Reduce > Combine values in a data set to create a new value >» Example: sum = (each elem in arr, total +=) reduce [1,2,3,4,5] returns 15 (the sum of the elements) Data data data data a data data data data data data data data data data data data desta data data data adata data data data data data data data data ata data data data data data data data data data adata Compute Cluster [0 i | [ Map | Map worker 1 iilap worker Map worker Input split Input split Input split Input flo parttioning 6 Fa eee [ki ky ket Kay] ty ky kv Ky) Kaw Way Wav ka I I T I Soran SO Soran ora e é Pantonng Ry Caer] 7 Synechronization ‘communication Re Way Hay Kav Fay av Riv Rov Ki [_Sarans 90a —] [eerendaoer | | song ans rousing aa ava ThnwRay Fig.6.5 Dataflow Implementation ba of MapReduce Sapa Caputo ouput Racice worker Reduce worker Reduce ee Ya ea ae A Simple Example a Counting words in a large set of documents eC une RCC) UR et ULL Jvalue: document contents for each word w in value Te eure ei an Beer mies men cere cK Ay plus an associated count of occurrences (just a “1” is recorded in this pseudo-code) eee CALC Cm CL) ) IIkey: word OU e co SRO oelE Ea eed counts emitted for a particular word aC ea Uy for each v in values result += Parselnt(v); Emit(AsString(result)); ‘Most people Intermediate (key, val) pairs nore most poetry (ignore, 1) Unique keys s ignores, 1 ® ignores (nore) —pf-=-—\ > [Gignore, (most, 1) (most. 1) Genores, 1) —op 1] (poetry. 1) (enost. 1) (vost, 1.11) (most.1) Sort (most. 1) Group 12,11 (poetty.1) (people, 1) ‘eee. tty j+~— ianores.1) (people. 1) (oety.331) Reduce” (poetry, 2 / (most, 1) (poetry, 1) (people, 1) (poetry, 1) MapReduce Most poetry library ignores most people rT Saad 18 mop task AGURE 6.6 YE ToM ere oe Rolle oR ece Mie MUR mole ei BUR esc R OB coo Ree ete ae Race D gu Cun hae cuus come ooo eeee Se sin tuna aC Relay R and the partitioning function are specified by the programmer. rr) MapReduce : Operation Steps When the user program calls the MapReduce function, the following sequence of actions occurs : 1) The MapReduce library in the user program first splits the input files into M pieces - 16 megabytes to 64 megabytes (MB) per piece. It then starts up many copies of program on a cluster of machines. 2) One of the copies of program is master. The rest are workers that are assigned work by the master. F) 20 MapReduce : Operation Steps 3) A worker who is assigned a map task : = reads the contents of the corresponding input split = parses key/value pairs out of the input data and passes each pair to the user - defined Map function. The intermediate key/value pairs produced by the Map function are buffered in memory. 4) The buffered pairs are written to local disk, partitioned into R regions by the partitioning function. The location of these buffered pairs on the local disk are passed back to the master, who forwards these locations to the reduce workers. Py] a MapReduce : Operation Steps 5) When a reduce worker is notified by the master about these locations, it reads the buffered data from the local lS ela la oR ele When a reduce worker has read all intermediate data, it SoMa NA MN Uefa ROM re 1a-] Melee let] of the same key are grouped together. 6) The reduce worker iterates over the sorted intermediate data and for each unique intermediate key, it passes the key and the corresponding set of intermediate values to the user's Reduce function. The output of the Reduce function is appended to a Lilar=|meleleol0l mil (om v2 Fr MapReduce : Operation Steps 7) When all map tasks and reduce tasks have been completed, the master wakes up the user program. At this point, MapReduce call in the user program returns back to the user code. After successful completion, output of the mapreduce execution is available in the R output files. 23 23 Logical Data Flow in 5 Processing Steps in MapReduce Process Intermediate (key, val) pairs Line offset Line content Unique keys (key!, (val, val. .J) ——of (keyS, val) (key8, val) Output Input MapReduce library ee Ya ea ae Py rey Locality issue e Master scheduling policy > Asks GFS for locations of replicas of input file blocks > Map tasks typically split into 64MB (== GFS block 742) > Map tasks scheduled so GFS input block replica are Co) gt aN MATA MOST AML 8L.6 e Effect >» Thousands of machines read input at local disk speed > Without this, rack switches limit read rate 25 Fault Tolerance e Reactive way Pan mca) - Heartbeat, Workers are periodically pinged by master « NO response = failed worker « If the processor of a worker fails, the tasks of that worker are reassigned to another worker. > Master failure = Master writes periodic checkpoints = Another master can be started from the last checkpointed Bei = If eventually the master dies, the job will be aborted 26 Fault Tolerance e Proactive way (Redundant Execution) oa The problem of “stragglers” (slow workers) = Other jobs consuming resources on machine = Bad disks with soft errors transfer data very slowly = Weird things: processor caches disabled (!!) When computation almost done, reschedule in- progress tasks Whenever either the primary or the backup executions finishes, mark it as completed rg Fault Tolerance e Input error: bad records > Map/Reduce functions sometimes fail for particular inputs Best solution is to debug & fix, but not always possible On segment fault = Send UDP packet to master from signal handler = Include sequence number of record being processed Skip bad records = If master sees two failures for same record, next worker is told to skip the record 28 Status monitor MapReduce status: MR_Indexer-beta6-large-2003_10_28 00_03 Fei Now 7 0951.07 2003 -- up Or 37 min 01 see ers, 1 death ‘Type _|Shards| Done| Active Input(MB)|Done(MB) | Output(MB) rl ‘Shulle Reduce! i i E calle Reduce shard ee 29 Points need to be emphasized e No reduce can begin until map is complete e Master must communicate locations of intermediate files e Tasks scheduled based on location of data e If map worker fails any time before reduce finishes, task must be completely rerun e MapReduce library does most of the hard work for us! Other Examples Distributed Grep: >» Map function emits a line if it matches a supplied pattern. » Reduce function is an identity function that copies the supplied intermediate data to the output. er ela O ae ee > Map function processes logs of web page requests and outputs , » Reduce function adds together all values for the same URL, emitting pairs. Reverse Web-Link graph; e.g., all URLs with reference to http://dblab.usc.edu: > Map function outputs for each link to a tgt in a page named src, >» Reduce concatenates the list of all src URLS associated with a given tgt URL and emits the pair: . Inverted Index; e.g., all URLs with 585 as a word: » Map function parses each document, emitting a sequence of , >» Reduce accepts all pairs for a given word, sorts the corresponding doc_IDs and emits a pair. > Set of all output pairs forms a simple inverted index. TT MapReduce Implementations elt i 1, Google 2, Apache Hadoop MapReduce Py SESE yy A if Yr I TT fete) el SO a Phoenix @ stanford GPU, Mars@HKUST Hadoop : software platform originally developed by Yahoo enabling users to write and run applications over vast distributed data. Attractive Features in Hadoop : = Scalable : can easily scale to store and process petabytes of data in the Web space = Economical : An open-source MapReduce minimizes the overheads in task spawning and massive data communication. | sion acs iea emai mea ke meted olse| across a large number of commodity nodes @ Reliable : Automatically maintains multiple copies of data to facilitate redeployment of computing tasks on failures Ce Ea ee ee Fr} Ped) <+— gigabit <— > I gigabit 40 nodes/rack, 1000-4000 nodes in cluster 1 Gbps bandwidth within rack, 8 Gbps out of rack Node specs (Yahoo terasort): 8 x 2GHz cores, 8 GB RAM, 4 disks (= 4 TB?) Challenges Cheap nodes fail, especially if you have many 1. Mean time between failures for 1 node = 3 years 2. Mean time between failures for 1000 nodes = 1 day 3. Solution: Build fault-tolerance into system Commodity network = low bandwidth 1. Solution: Push computation to the data Programming distributed systems is hard 1. Solution: Data-parallel programming model: users write “map” & “reduce” functions, system distributes aCe) a= Lao areal (xm cole Les) cry Hadoop Components e Distributed file system (HDFS) > Single namespace for entire cluster >» Replicates data 3x for fault-tolerance e MapReduce framework > Executes user jobs specified as “map” and “reduce” functions >» Manages work distribution & fault-tolerance cid Hadoop Distributed File System e Files split into 128MB Iollele ea Namenode e Blocks replicated across Ler several datanodes (usually Kc) e Single namenode stores metadata (file names, block cE E& P| locations, etc) _ a e Optimized for large files, Eyeto (0 (vate 1 mace le se Yen zel stele) e Files are append-only Cluster > 4 JobTracker MapReduce| engnie NameNode FIGURE 6.11 HDFS and MapReduce architecture in Hadoop. Ss Secure Query Processing with Hadoop/MapReduce Se es a a a 7) Higher-level languages over Hadoop: Pig and Hive Motivation e Many parallel algorithms can be expressed by a series of MapReduce jobs e But MapReduce is fairly low-level: must think about keys, values, partitioning, etc e Can we capture common “job building blocks”? Pig Started at Yahoo! Research Runs about 30% of Yahoo!’s jobs mere lK os 4 2 > Expresses sequences of MapReduce jobs Data model: nested “bags” of items Provides relational (SQL) operators (JOIN, GROUP BY, etc) Easy to plug in Java functions Pig Pen development environment for Eclipse An Example Problem Suppose you have user data in one file, page view data in another, and you need to find the top 5 most visited pages by users aged 18 - 25. ee ene Th ry Lee ne ented In MapReduce In Pig Latin Users Lani -Ta-10) Py Joined Grouped Summed Sorted Top5S Ee 001-1 |=] <-D See elon Ut 1a ob age >= 18 ©) age <= 25; = load ‘pages’ as (user, url); = ©.) Filtered >. name, Pages ©. user; EamenmenerOnmns (01 11-10 MORIN [an ed Serene C1 0) ] (10 MCU SUSRMORE 401 0)0] 09 coun’ (Joined) >= clicks; EenmerenammsY©((|1|(-1¢ ORIN cn OK cL 4ST = 9.— Sorted 5; store TopS into ‘topSsites’; ry ee eee ee LT ee MeL mente tite Ease of Translation Notice how naturally the components of the job translate into Pig Latin. Users = (000 .. Lis ho el(e oY -4-5 ee Joined = ©... eK = group ... i T0111 (=10 eo @ Po sorted order .. TopS = limit .. — Ca ee ee ee ee LT ee MeL nem tee iad Ease of Translation Notice how naturally the components of the job translate into Pig Latin. *Users = load ... +Filtered = filter .. Pages = load ... Joined = ©... .——— Grouped = group .. _—Summed = .. (00 (+). ~ Sorted = order .. Da eee RI) 4 Job 3 Cee ee ae aoa oe ee ee im INVA) Developed at Facebook Used for majority of Facebook jobs “Relational database” built on Hadoop Maintains list of table schemas SQL-like query language (HQL) Can call Hadoop Streaming scripts from HQL Supports table partitioning, clustering, complex data types, some optimizations ana Na A Sample Hive Queries + Find top 5 pages visited by users aged 18-25: SELECT p.url, COUNT(1) as clicks FROM users u JOIN page views p ON (u.name = p.user) WHERE u.age >= 18 AND u.age <= 25 GROUP BY p.url ORDER BY clicks te es + Filter page views through Python script: SELECT TRANSFORM(p.user, p.date) USING ‘map_script.py’ AS dt, uid CLUSTER BY dt FROM page_views p; 7) Amazon Elastic MapReduce e Provides a web-based interface and command- line tools for running Hadoop jobs on Amazon |= 072 Data stored in Amazon S3 Monitors job and shuts down machines after use e Small extra charge on top of EC2 pricing e If you want more control over how you Hadoop runs, you can launch a Hadoop cluster on EC2 manually using the scripts in src/contrib/ec2 or Elastic MapReduce Workflow Create a New Job Flow Creating a job low to process your data using Amazon Elastic MapReduce is simple and quick. Let's begin by giving your Job low a name: land selecting ts type. If you don't already have an application you'd ike to run on Amazon Elastic MapReduce, samples are availabe to help you get stated. Job Flow Name*: ly Job Flow Types: @ Streaming O custom Jar O Pig Program Q Sample Applications Wor our Svaning” Word cur se Pinon ppeaton tht cours ecureces lech nord Elastic MapReduce Workflow Create a New Job Flow Specty Mapper and Reducer functions to run within the ob Flow. The mapper and reducers may be ether (1) ass names refering toa mapper or reducer cass in Hadoop or (i) locations in Amazon $3. (Ck Mere fora list of avaiable tools to help you upload and download fle from Amazon S3.) The format for spectying aloction in Amazon $3 ls bucket-name/path. name. Te locton should pont to an, ‘executable program, for example a python program. Extra arguments are passed tthe Hadoop streaming program and ca specty things Such as adaiona fas tobe loaded into the dtibuted cache Input Location*: ‘laslcmapreducelsamploslwordcounVinpat ‘Output Location: '/wordcourl otpu/2008-08-79 Mapper*: lasticmapreducelsamplesiworcourtiworSpiier py Reducer*: aggregate extra Args: Create a New Job Flow Enter the number and typeof E You'd tke to run your Job flow on. Number of instances: ‘Type of Instance: (Snail i cy ‘Amazon ec2 | Sispmegues —_Cbudront Conclusions MapReduce programming model hides the complexity of work distribution and fault tolerance Principal design philosophies: > Make it scalable, so you can throw hardware at problems > Make it cheap, lowering hardware, programming and admin costs MapReduce is not suitable for all problems, but when it works, it may save you quite a bit of time Cloud computing makes it straightforward to start using Hadoop (or other parallel software) at scale cy Resources Hadoop: http://hadoop.apache.org/core/ Pig: http://hadoop.apache.org/pig Hive: http://hadoop.apache.org/hive Video tutorials: http://www.cloudera.com/hadoop-training Amazon Web Services: http://aws.amazon.com/ Amazon Elastic MapReduce guide: http://docs.amazonwebservices.com/ElasticMapReduce/lat est/GettingStartedGuide/ cg

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy