Presentation on theme: "Fundamentals of MapReduce Jimmy Lin The iSchool University of Maryland Monday, March 30, 2009 This work is licensed under a Creative Commons Attribution-Noncommercial-Share."— Presentation transcript:
Fundamentals of MapReduce Jimmy Lin The iSchool University of Maryland Monday, March 30, 2009 This work is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 3.0 United States See http://creativecommons.org/licenses/by-nc-sa/3.0/us/ for details Some material adapted from slides by Christophe Bisciglia, Aaron Kimball, & Sierra Michels-Slettvet, Google Distributed Computing Seminar, 2007 (licensed under Creation Commons Attribution 3.0 License)
This afternoon... Introduction to MapReduce MapReduce “killer app” #1: text retrieval MapReduce “killer app” #2: graph algorithms
Introduction to MapReduce: Topics Functional programming MapReduce Distributed file system
Functional Programming MapReduce = functional programming meets distributed processing on steroids Not a new idea… dates back to the 50’s (or even 30’s) What is functional programming? Computation as application of functions Theoretical foundation provided by lambda calculus How is it different? Traditional notions of “data” and “instructions” are not applicable Data flows are implicit in program Different orders of execution are possible Exemplified by LISP and ML
Overview of Lisp Lisp ≠ Lost In Silly Parentheses We’ll focus on particular a dialect: “Scheme” Lists are primitive data types Functions written in prefix notation (+ 1 2) 3 (* 3 4) 12 (sqrt (+ (* 3 3) (* 4 4))) 5 (define x 3) x (* x 5) 15 '(1 2 3 4 5) '((a 1) (b 2) (c 3))
Functions Functions = lambda expressions bound to variables Syntactic sugar for defining functions Above expressions is equivalent to: Once defined, function can be applied: (define (foo x y) (sqrt (+ (* x x) (* y y)))) (define foo (lambda (x y) (sqrt (+ (* x x) (* y y))))) (foo 3 4) 5
Other Features In Scheme, everything is an s-expression No distinction between “data” and “code” Easy to write self-modifying code Higher-order functions Functions that take other functions as arguments (define (bar f x) (f (f x))) (define (baz x) (* x x)) (bar baz 2) 16 Doesn’t matter what f is, just apply it twice.
Recursion is your friend Simple factorial example Even iteration is written with recursive calls! (define (factorial n) (if (= n 1) 1 (* n (factorial (- n 1))))) (factorial 6) 720 (define (factorial-iter n) (define (aux n top product) (if (= n top) (* n product) (aux (+ n 1) top (* n product)))) (aux 1 n 1)) (factorial-iter 6) 720
Lisp MapReduce? What does this have to do with MapReduce? After all, Lisp is about processing lists Two important concepts in functional programming Map: do something to everything in a list Fold: combine results of a list in some way
Map Map is a higher-order function How map works: Function is applied to every element in a list Result is a new list fffff
Fold Fold is also a higher-order function How fold works: Accumulator set to initial value Function applied to list element and the accumulator Result stored in the accumulator Repeated for every item in the list Result is the final value in the accumulator ffffffinal value Initial value
Lisp MapReduce Let’s assume a long list of records: imagine if... We can parallelize map operations We have a mechanism for bringing map results back together in the fold operation That’s MapReduce! Observations: No limit to map parallelization since maps are indepedent We can reorder folding if the fold function is commutative and associative
Typical Problem Iterate over a large number of records Extract something of interest from each Shuffle and sort intermediate results Aggregate intermediate results Generate final output Key idea: provide an abstraction at the point of these two operations Map Reduce
MapReduce Programmers specify two functions: map (k, v) → * reduce (k’, v’) → * All v’ with the same k’ are reduced together Usually, programmers also specify: partition (k’, number of partitions ) → partition for k’ Often a simple hash of the key, e.g. hash(k’) mod n Allows reduce operations for different keys in parallel Implementations: Google has a proprietary implementation in C++ Hadoop is an open source implementation in Java
Recall these problems? How do we assign work units to workers? What if we have more work units than workers? What if workers need to share partial results? How do we aggregate partial results? How do we know all the workers have finished? What if workers die?
MapReduce Runtime Handles scheduling Assigns workers to map and reduce tasks Handles “data distribution” Moves the process to the data Handles synchronization Gathers, sorts, and shuffles intermediate data Handles faults Detects worker failures and restarts Everything happens on top of a distributed FS (later)
“Hello World”: Word Count Map(String input_key, String input_value): // input_key: document name // input_value: document contents for each word w in input_values: EmitIntermediate(w, "1"); Reduce(String key, Iterator intermediate_values): // key: a word, same for input and output // intermediate_values: a list of counts int result = 0; for each v in intermediate_values: result += ParseInt(v); Emit(AsString(result));
split 0 split 1 split 2 split 3 split 4 worker Master User Program output file 0 output file 1 (1) fork (2) assign map (2) assign reduce (3) read (4) local write (5) remote read (6) write Input files Map phase Intermediate files (on local disk) Reduce phase Output files Redrawn from Dean and Ghemawat (OSDI 2004)
Bandwidth Optimization Issue: large number of key-value pairs Solution: use “Combiner” functions Executed on same machine as mapper Results in a “mini-reduce” right after the map phase Reduces key-value pairs to save bandwidth
Skew Problem Issue: reduce is only as fast as the slowest map Solution: redundantly execute map operations, use results of first to finish Addresses hardware problems... But not issues related to inherent distribution of data
How do we get data to the workers? Compute Nodes NAS SAN What’s the problem here?
Distributed File System Don’t move data to workers… Move workers to the data! Store data on the local disks for nodes in the cluster Start up the workers on the node that has the data local Why? Not enough RAM to hold all the data in memory Disk access is slow, disk throughput is good A distributed file system is the answer GFS (Google File System) HDFS for Hadoop
GFS: Assumptions Commodity hardware over “exotic” hardware High component failure rates Inexpensive commodity components fail all the time “Modest” number of HUGE files Files are write-once, mostly appended to Perhaps concurrently Large streaming reads over random access High sustained throughput over low latency GFS slides adapted from material by Dean et al.
GFS: Design Decisions Files stored as chunks Fixed size (64MB) Reliability through replication Each chunk replicated across 3+ chunkservers Single master to coordinate access, keep metadata Simple centralized management No data caching Little benefit due to large data sets, streaming reads Simplify the API Push some of the issues onto the client
Redrawn from Ghemawat et al. (SOSP 2003) Application GSF Client GFS master File namespace /foo/bar chunk 2ef0 GFS chunkserver Linux file system … GFS chunkserver Linux file system … (file name, chunk index) (chunk handle, chunk location) Instructions to chunkserver Chunkserver state (chunk handle, byte range) chunk data
Single Master We know this is a: Single point of failure Scalability bottleneck GFS solutions: Shadow masters Minimize master involvement Never move data through it, use only for metadata (and cache metadata at clients) Large chunk size Master delegates authority to primary replicas in data mutations (chunk leases) Simple, and good enough!
Master’s Responsibilities (1/2) Metadata storage Namespace management/locking Periodic communication with chunkservers Give instructions, collect state, track cluster health Chunk creation, re-replication, rebalancing Balance space utilization and access speed Spread replicas across racks to reduce correlated failures Re-replicate data if redundancy falls below threshold Rebalance data to smooth out storage and request load
Master’s Responsibilities (2/2) Garbage Collection Simpler, more reliable than traditional file delete Master logs the deletion, renames the file to a hidden name Lazily garbage collects hidden files Stale replica deletion Detect “stale” replicas using chunk version numbers
Metadata Global metadata is stored on the master File and chunk namespaces Mapping from files to chunks Locations of each chunk’s replicas All in memory (64 bytes / chunk) Fast Easily accessible Master has an operation log for persistent logging of critical metadata updates Persistent on local disk Replicated Checkpoints for faster recovery
Mutations Mutation = write or append Must be done for all replicas Goal: minimize master involvement Lease mechanism: Master picks one replica as primary; gives it a “lease” for mutations Primary defines a serial order of mutations All replicas follow this order Data flow decoupled from control flow
Parallelization Problems How do we assign work units to workers? What if we have more work units than workers? What if workers need to share partial results? How do we aggregate partial results? How do we know all the workers have finished? What if workers die? How is MapReduce different?
Text Retrieval: Topics Introduction to information retrieval (IR) Boolean retrieval Ranked retrieval Text retrieval with MapReduce
The Information Retrieval Cycle Source Selection Source Selection Search Query Selection Results Examination Documents Delivery Information Query Formulation Query Formulation Resource source reselection System discovery Vocabulary discovery Concept discovery Document discovery
The Central Problem in Search Searcher Author Concepts Query Terms Document Terms Do these represent the same concepts? “tragic love story”“fateful star-crossed romance”
Architecture of IR Systems Documents Query Hits Representation Function Representation Function Query RepresentationDocument Representation Comparison Function Index offlineonline
How do we represent text? Remember: computers don’t “understand” anything! “Bag of words” Treat all the words in a document as index terms for that document Assign a “weight” to each term based on “importance” Disregard order, structure, meaning, etc. of the words Simple, yet effective! Assumptions Term occurrence is independent Document relevance is independent “Words” are well-defined
What’s a word? 天主教教宗若望保祿二世因感冒再度住進醫院。 這是他今年第二度因同樣的病因住院。 وقال مارك ريجيف - الناطق باسم الخارجية الإسرائيلية - إن شارون قبل الدعوة وسيقوم للمرة الأولى بزيارة تونس، التي كانت لفترة طويلة المقر الرسمي لمنظمة التحرير الفلسطينية بعد خروجها من لبنان عام 1982. Выступая в Мещанском суде Москвы экс-глава ЮКОСа заявил не совершал ничего противозаконного, в чем обвиняет его генпрокуратура России. भारत सरकार ने आर्थिक सर्वेक्षण में वित्तीय वर्ष 2005-06 में सात फ़ीसदी विकास दर हासिल करने का आकलन किया है और कर सुधार पर ज़ोर दिया है 日米連合で台頭中国に対処 … アーミテージ前副長官提言 조재영 기자 = 서울시는 25 일 이명박 시장이 ` 행정중심복합도시 '' 건설안 에 대해 ` 군대라도 동원해 막고싶은 심정 '' 이라고 말했다는 일부 언론의 보도를 부인했다.
Sample Document McDonald's slims down spuds Fast-food chain to reduce certain types of fat in its french fries with new cooking oil. NEW YORK (CNN/Money) - McDonald's Corp. is cutting the amount of "bad" fat in its french fries nearly in half, the fast-food chain said Tuesday as it moves to make all its fried menu items healthier. But does that mean the popular shoestring fries won't taste the same? The company says no. "It's a win-win for our customers because they are getting the same great french-fry taste along with an even healthier nutrition profile," said Mike Roberts, president of McDonald's USA. But others are not so sure. McDonald's will not specifically discuss the kind of oil it plans to use, but at least one nutrition expert says playing with the formula could mean a different taste. Shares of Oak Brook, Ill.-based McDonald's (MCD: down $0.54 to $23.22, Research, Estimates) were lower Tuesday afternoon. It was unclear Tuesday whether competitors Burger King and Wendy's International (WEN: down $0.80 to $34.91, Research, Estimates) would follow suit. Neither company could immediately be reached for comment. … 16 × said 14 × McDonalds 12 × fat 11 × fries 8 × new 6 × company, french, nutrition 5 × food, oil, percent, reduce, taste, Tuesday … “Bag of Words”
Boolean Retrieval Users express queries as a Boolean expression AND, OR, NOT Can be arbitrarily nested Retrieval is based on the notion of sets Any given query divides the collection into two sets: retrieved, not-retrieved Pure Boolean systems do not define an ordering of the results
Representing Documents The quick brown fox jumped over the lazy dog’s back. Document 1 Document 2 Now is the time for all good men to come to the aid of their party. the is for to of quick brown fox over lazy dog back now time all good men come jump aid their party 0 0 1 1 0 1 1 0 1 1 0 0 1 0 1 0 0 1 1 0 0 1 0 0 1 0 0 1 1 0 1 0 1 1 Term Document 1Document 2 Stopword List
Inverted Index quick brown fox over lazy dog back now time all good men come jump aid their party 48 246 137 1357 2468 35 357 2468 3 1357 13578 248 268 157 246 13 68 Term Postings
Boolean Retrieval To execute a Boolean query: Build query syntax tree For each clause, look up postings Traverse postings and apply Boolean operator Efficiency analysis Postings traversal is linear (assuming sorted postings) Start with shortest posting first ( fox or dog ) and quick foxdog ORquick AND fox dog35 357 fox dog35 357 OR = union 357
Extensions Implementing proximity operators Store word offset in postings Handling term variations Stem words: love, loving, loves … lov
Strengths and Weaknesses Strengths Precise, if you know the right strategies Precise, if you have an idea of what you’re looking for Implementations are fast and efficient Weaknesses Users must learn Boolean logic Boolean logic insufficient to capture the richness of language No control over size of result set: either too many hits or none When do you stop reading? All documents in the result set are considered “equally good” What about partial matches? Documents that “don’t quite match” the query may be useful also
Ranked Retrieval Order documents by how likely they are to be relevant to the information need Estimate relevance(q, d i ) Sort documents by relevance Display sorted results User model Present hits one screen at a time, best results first At any point, users can decide to stop looking How do we estimate relevance? Assume document is relevant if it has a lot of query terms Replace relevance (q, d i ) with sim(q, d i ) Compute similarity of vector representations
Vector Space Model Assumption: Documents that are “close together” in vector space “talk about” the same things t1t1 d2d2 d1d1 d3d3 d4d4 d5d5 t3t3 t2t2 θ φ Therefore, retrieve documents based on how close the document is to the query (i.e., similarity ~ “closeness”)
Similarity Metric How about |d 1 – d 2 |? Instead of Euclidean distance, use “angle” between the vectors It all boils down to the inner product (dot product) of vectors
Term Weighting Term weights consist of two components Local: how important is the term in this document? Global: how important is the term in the collection? Here’s the intuition: Terms that appear often in a document should get high weights Terms that appear in many documents should get low weights How do we capture this mathematically? Term frequency (local) Inverse document frequency (global)
TF.IDF Term Weighting weight assigned to term i in document j number of occurrence of term i in document j number of documents in entire collection number of documents with term i
Sketch: Scoring Algorithm Initialize accumulators to hold document scores For each query term t in the user’s query Fetch t’s postings For each document, score doc += w t,d w t,q Apply length normalization to the scores at end Return top N documents
MapReduce it? The indexing problem Must be relatively fast, but need not be real time For Web, incremental updates are important The retrieval problem Must have sub-second response For Web, only need relatively few results
Indexing: Performance Analysis Inverted indexing is fundamental to all IR models Fundamentally, a large sorting problem Terms usually fit in memory Postings usually don’t How is it done on a single machine? How large is the inverted index? Size of vocabulary Size of postings
Vocabulary Size: Heaps’ Law V is vocabulary size n is corpus size (number of documents) K and are constants Typically, K is between 10 and 100, is between 0.4 and 0.6 When adding new documents, the system is likely to have seen most terms already… but the postings keep growing
George Kingsley Zipf (1902-1950) observed the following relation between frequency and rank In other words: A few elements occur very frequently Many elements occur very infrequently Zipfian distributions: English words Library book checkout patterns Website popularity (almost anything on the Web) Postings Size: Zipf’s Law or f = frequency r = rank c = constant
Word Frequency in English Frequency of 50 most common words in English (sample of 19 million words)
Does it fit Zipf’s Law? The following shows rf 1000 r is the rank of word w in the sample f is the frequency of word w in the sample
MapReduce: Index Construction Map over all documents Emit term as key, (docid, tf) as value Emit other information as necessary (e.g., term position) Reduce Trivial: each value represents a posting! Might want to sort the postings (e.g., by docid or tf) MapReduce does all the heavy lifting!
Query Execution MapReduce is meant for large-data batch processing Not suitable for lots of real time operations requiring low latency The solution: “the secret sauce” Most likely involves document partitioning Lots of system engineering: e.g., caching, load balancing, etc.
MapReduce: Query Execution High-throughput batch query execution: Instead of sequentially accumulating scores per query term: Have mappers traverse postings in parallel, emitting partial score components Reducers serve as the accumulators, summing contributions for each query term MapReduce does all the heavy lifting Replace random access with sequential reads Amortize over lots of queries Examine multiple postings in parallel
Graph Algorithms: Topics Introduction to graph algorithms and graph representations Single Source Shortest Path (SSSP) problem Refresher: Dijkstra’s algorithm Breadth-First Search with MapReduce PageRank
What’s a graph? G = (V,E), where V represents the set of vertices (nodes) E represents the set of edges (links) Both vertices and edges may contain additional information Different types of graphs: Directed vs. undirected edges Presence or absence of cycles Graphs are everywhere: Hyperlink structure of the Web Physical structure of computers on the Internet Interstate highway system Social networks
Some Graph Problems Finding shortest paths Routing Internet traffic and UPS trucks Finding minimum spanning trees Telco laying down fiber Finding Max Flow Airline scheduling Identify “special” nodes and communities Breaking up terrorist cells, spread of avian flu Bipartite matching Monster.com, Match.com And of course... PageRank
Graphs and MapReduce Graph algorithms typically involve: Performing computation at each node Processing node-specific data, edge-specific data, and link structure Traversing the graph in some manner Key questions: How do you represent graph data in MapReduce? How do you traverse a graph in MapReduce?
Representing Graphs G = (V, E) A poor representation for computational purposes Two common representations Adjacency matrix Adjacency list
Adjacency Matrices Represent a graph as an n x n square matrix M n = |V| M ij = 1 means a link from node i to j 1234 10101 21011 31000 41010 1 1 2 2 3 3 4 4
Adjacency Matrices: Critique Advantages: Naturally encapsulates iteration over nodes Rows and columns correspond to inlinks and outlinks Disadvantages: Lots of zeros for sparse matrices Lots of wasted space
Adjacency Lists Take adjacency matrices… and throw away all the zeros 1234 10101 21011 31000 41010 1: 2, 4 2: 1, 3, 4 3: 1 4: 1, 3
Adjacency Lists: Critique Advantages: Much more compact representation Easy to compute over outlinks Graph structure can be broken up and distributed Disadvantages: Much more difficult to compute over inlinks
Single Source Shortest Path Problem: find shortest path from a source node to one or more target nodes First, a refresher: Dijkstra’s Algorithm
Dijkstra’s Algorithm Example 0 0 10 5 23 2 1 9 7 46 Example from CLR
Dijkstra’s Algorithm Example 0 0 10 5 5 5 23 2 1 9 7 46 Example from CLR
Dijkstra’s Algorithm Example 0 0 8 5 5 14 7 7 10 5 23 2 1 9 7 46 Example from CLR
Dijkstra’s Algorithm Example 0 0 8 8 5 5 13 7 7 10 5 23 2 1 9 7 46 Example from CLR
Dijkstra’s Algorithm Example 0 0 8 8 5 5 9 9 7 7 10 5 23 2 1 9 7 46 Example from CLR
Dijkstra’s Algorithm Example 0 0 8 8 5 5 9 9 7 7 10 5 23 2 1 9 7 46 Example from CLR
Single Source Shortest Path Problem: find shortest path from a source node to one or more target nodes Single processor machine: Dijkstra’s Algorithm MapReduce: parallel Breadth-First Search (BFS)
Finding the Shortest Path First, consider equal edge weights Solution to the problem can be defined inductively Here’s the intuition: DistanceTo(startNode) = 0 For all nodes n directly reachable from startNode, DistanceTo(n) = 1 For all nodes n reachable from some other set of nodes S, DistanceTo(n) = 1 + min(DistanceTo(m), m S)
From Intuition to Algorithm A map task receives Key: node n Value: D (distance from start), points-to (list of nodes reachable from n) p points-to: emit (p, D+1) The reduce task gathers possible distances to a given p and selects the minimum one
Multiple Iterations Needed This MapReduce task advances the “known frontier” by one hop Subsequent iterations include more reachable nodes as frontier advances Multiple iterations are needed to explore entire graph Feed output back into the same MapReduce task Preserving graph structure: Problem: Where did the points-to list go? Solution: Mapper emits (n, points-to) as well
Termination Does the algorithm ever terminate? Eventually, all nodes will be discovered, all edges will be considered (in a connected graph) When do we stop?
Weighted Edges Now add positive weights to the edges Simple change: points-to list in map task includes a weight w for each pointed-to node emit (p, D+w p ) instead of (p, D+1) for each node p Does this ever terminate? Yes! Eventually, no better distances will be found. When distance is the same, we stop Mapper should emit (n, D) to ensure that “current distance” is carried into the reducer
Comparison to Dijkstra Dijkstra’s algorithm is more efficient At any step it only pursues edges from the minimum-cost path inside the frontier MapReduce explores all paths in parallel Divide and conquer Throw more hardware at the problem
General Approach MapReduce is adapt at manipulating graphs Store graphs as adjacency lists Graph algorithms with for MapReduce: Each map task receives a node and its outlinks Map task compute some function of the link structure, emits value with target as the key Reduce task collects keys (target nodes) and aggregates Iterate multiple MapReduce cycles until some termination condition Remember to “pass” graph structure from one iteration to next
Random Walks Over the Web Model: User starts at a random Web page User randomly clicks on links, surfing from page to page What’s the amount of time that will be spent on any given page? This is PageRank
Given page x with in-bound links t 1 …t n, where C(t) is the out-degree of t is probability of random jump N is the total number of nodes in the graph PageRank: Defined X X t1t1 t1t1 t2t2 t2t2 tntn tntn …
Computing PageRank Properties of PageRank Can be computed iteratively Effects at each iteration is local Sketch of algorithm: Start with seed PR i values Each page distributes PR i “credit” to all pages it links to Each target page adds up “credit” from multiple in-bound links to compute PR i+1 Iterate until values converge
PageRank in MapReduce Map: distribute PageRank “credit” to link targets... Reduce: gather up PageRank “credit” from multiple sources to compute new PageRank value Iterate until convergence
PageRank: Issues Is PageRank guaranteed to converge? How quickly? What is the “correct” value of , and how sensitive is the algorithm to it? What about dangling links? How do you know when to stop?