Distributed Computations MapReduce adapted from Jeff Dean’s slides.
Basics of Cloud Computing –Lecture 3 Introduction to MapReduce · map cannot see the effects of...
Transcript of Basics of Cloud Computing –Lecture 3 Introduction to MapReduce · map cannot see the effects of...
Basics of Cloud Computing – Lecture 3
Introduction to MapReduce
Satish Srirama
Some material adapted from slides by Jimmy Lin, Christophe Bisciglia, Aaron Kimball,
& Sierra Michels-Slettvet, Google Distributed Computing Seminar, 2007 (licensed
under Creation Commons Attribution 3.0 License)
Outline
• Functional programming review
• MapReduce
• Hadoop
• HDFS (Hadoop Distributed File System)• HDFS (Hadoop Distributed File System)
18.04.2012 2/39Satish Srirama
Economics of Cloud Providers –
Failures - Recap
• Cloud Computing providers bring a shift from high reliability/availability servers to commodity servers– At least one failure per day in large datacenter
• Caveat: User software has to adapt to failures• Caveat: User software has to adapt to failures
• Solution: Replicate data and computation– MapReduce & Distributed File System
• MapReduce = functional programming meets distributed processing on steroids – Not a new idea… dates back to the 50’s (or even 30’s)
18.04.2012 3/39Satish Srirama
Functional programming
• What is functional programming?
– Computation as application of functions
– Theoretical foundation provided by lambda calculus
• How is it different?• How is it different?
– Traditional notions of “data” and “instructions” are
not applicable
– Data flows are implicit in program design
– Different orders of execution are possible
• Exemplified by LISP and ML
18.04.2012 4/39Satish Srirama
Functional Programming Review
• Lists are primitive data types
• Functions = lambda expressions bound to variables
– f = lambda x : 2 * x– f = lambda x : 2 * x
print f(2)
• Higher-order functions
– Functions that take other functions as arguments
• Recursion is your friend
18.04.2012 5/39Satish Srirama
Functional Programming - features
• Functional operations do not modify data structures: They always create new ones
• Original data still exists in unmodified form
• Data flows are implicit in program design• Data flows are implicit in program design
• Order of operations does not matter
18.04.2012 6/39Satish Srirama
FP features - continued
• fun foo(l: int list) =
sum(l) + mul(l) + length(l)– Order of sum() and mul(), etc does not matter – they do not
modify l
• Functional Updates Do Not Modify Structures– fun append(x, lst) = – fun append(x, lst) =
let lst' = reverse lst in
reverse ( x :: lst' )
– The append() function reverses a list, adds a new element to the front, and returns all of that, reversed, which appends an item.
– But it never modifies lst!
• Functions Can Be Used As Arguments– fun DoDouble(f, x) = f (f x)
18.04.2012 7/39Satish Srirama
Functional Programming ->
MapReduce
• Two important concepts in functional
programming
– Map: do something to everything in a list
– Fold: combine results of a list in some way– Fold: combine results of a list in some way
18.04.2012 8/39Satish Srirama
Map
• Map is a higher-order function
• How map works:– Function is applied to every element in a list
– Result is a new list
• map f lst: (’a->’b) -> (’a list) -> (’b list)• map f lst: (’a->’b) -> (’a list) -> (’b list)– Creates a new list by applying f to each element of the
input list; returns output in order.
18.04.2012
f f f f f f
9/39Satish Srirama
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– Result stored in the accumulator
– Repeated for every item in the list
– Result is the final value in the accumulator
18.04.2012 10/39Satish Srirama
Map/Fold in Action
• Simple map example:
• Fold examples:
(map (lambda (x) (* x x))
'(1 2 3 4 5))
→ '(1 4 9 16 25)
• Fold examples:
• Sum of squares:
(fold + 0 '(1 2 3 4 5)) → 15
(fold * 1 '(1 2 3 4 5)) → 120
(define (sum-of-squares v)
(fold + 0 (map (lambda (x) (* x x)) v)))
(sum-of-squares '(1 2 3 4 5)) → 55
18.04.2012 11/39Satish Srirama
Implicit Parallelism In map
• In a purely functional setting, elements of a list being computed by map cannot see the effects of the computations on other elements
• If order of application of f to elements in list is commutative, we can reorder or parallelize execution
• 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 – We have a mechanism for bringing map results back together in the fold operation
• This is the “secret” that MapReduce exploits
• Observations:– No limit to map parallelization since maps are independent
– We can reorder folding if the fold function is commutative and associative
18.04.2012 12/39Satish Srirama
Typical Large-Data Problem
• Iterate over a large number of records
• Extract something of interest from each
• Shuffle and sort intermediate results
• Aggregate intermediate results• Aggregate intermediate results
• Generate final outputKey idea: provide a functional abstraction for
these two operations – MapReduce
18.04.2012 13/39Satish Srirama
MapReduce
• Programmers specify two functions:
map (k, v) → <k’, v’>*
reduce (k’, v’) → <k’, v’>*
– All values with the same key are sent to the same – All values with the same key are sent to the same reducer
• The execution framework handles everything else…
18.04.2012 14/39Satish Srirama
mapmap map map
k1 k2 k3 k4 k5 k6v1 v2 v3 v4 v5 v6
ba 1 2 c c3 6 a c5 2 b c7 8
MapReduce
Shuffle and Sort: aggregate values by keys
reduce reduce reduce
ba 1 2 c c3 6 a c5 2 b c7 8
a 1 5 b 2 7 c 2 3 6 8
r1 s1 r2 s2 r3 s3
18.04.2012 15Satish Srirama
“Hello World” Example: Word Count
Map(String docid, String text):
for each word w in text:
Emit(w, 1);
Reduce(String term, Iterator<Int> values):
int sum = 0;
for each v in values:
sum += v;
Emit(term, sum);
18.04.2012 16/39Satish Srirama
MapReduce
• Programmers specify two functions:
map (k, v) → <k’, v’>*
reduce (k’, v’) → <k’, v’>*
– All values with the same key are sent to the same – All values with the same key are sent to the same reducer
• The execution framework handles everything else…
What’s “everything else”?
18.04.2012 17/39Satish Srirama
MapReduce “Runtime”• Handles scheduling
– Assigns workers to map and reduce tasks
• Handles “data distribution”– Moves processes to data
• Handles synchronizationGathers, sorts, and shuffles intermediate data– Gathers, sorts, and shuffles intermediate data
• Handles errors and faults– Detects worker failures and automatically restarts
• Handles speculative execution– Detects “slow” workers and re-executes work
• Everything happens on top of a distributed FS (later)
Sounds simple, but many challenges!18.04.2012 18/39Satish Srirama
MapReduce - extended
• Programmers specify two functions:map (k, v) → <k’, v’>*reduce (k’, v’) → <k’, v’>*– All values with the same key are reduced together
• The execution framework handles everything else…• Not quite…usually, programmers also specify:• Not quite…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– Divides up key space for parallel reduce operationscombine (k’, v’) → <k’, v’>*– Mini-reducers that run in memory after the map phase– Used as an optimization to reduce network traffic
18.04.2012 19/39Satish Srirama
combinecombine combine combine
ba 1 2 c 9 a c5 2 b c7 8
mapmap map map
k1 k2 k3 k4 k5 k6v1 v2 v3 v4 v5 v6
ba 1 2 c c3 6 a c5 2 b c7 8
ba 1 2 c 9 a c5 2 b c7 8
partition partition partition partition
Shuffle and Sort: aggregate values by keys
reduce reduce reduce
a 1 5 b 2 7 c 2 9 8
r1 s1 r2 s2 r3 s3
c 2 3 6 8
18.04.2012 20Satish Srirama
Two more details…
• Barrier between map and reduce phases
– But we can begin copying intermediate data
earlier
• Keys arrive at each reducer in sorted order• Keys arrive at each reducer in sorted order
– No enforced ordering across reducers
18.04.2012 21/39Satish Srirama
split 0
worker
Master
User
Program
output
(1) submit
(2) schedule map (2) schedule reduce
MapReduce Overall Architecture
split 0
split 1
split 2
split 3
split 4
worker
worker
worker
worker
output
file 0
output
file 1
(3) read(4) local write
(5) remote read (6) write
Input
files
Map
phase
Intermediate files
(on local disk)
Reduce
phase
Output
files
Adapted from (Dean and Ghemawat, OSDI 2004)
18.04.2012 22Satish Srirama
MapReduce can refer to…
• The programming model
• The execution framework (aka “runtime”)
• The specific implementation
Usage is usually clear from context!
18.04.2012 23/39Satish Srirama
MapReduce Implementations
• Google has a proprietary implementation in C++
– Bindings in Java, Python
• Hadoop is an open-source implementation in Java
– Development led by Yahoo, used in production– Development led by Yahoo, used in production
– Now an Apache project
– Rapidly expanding software ecosystem, but still lots of
room for improvement (e.g., OSDI 2008, Nexus)
• Lots of custom research implementations
– For GPUs, cell processors, etc.
18.04.2012 24/39Satish Srirama
Cloud Computing Storage, or how do
we get data to the workers?NAS
Compute Nodes
SAN
What’s the problem here?
18.04.2012 25/39Satish Srirama
Distributed File System
• Don’t move data to workers… move workers to the data!– Store data on the local disks of nodes in the cluster
– Start up the workers on the node that has the data local
• Why?• Why?– Network bisection bandwidth is limited
– Not enough RAM to hold all the data in memory
– Disk access is slow, but disk throughput is reasonable
• A distributed file system is the answer– GFS (Google File System) for Google’s MapReduce
– HDFS (Hadoop Distributed File System) for Hadoop
18.04.2012 26/39Satish Srirama
GFS: Assumptions
• Choose commodity hardware over “exotic” hardware
– Scale “out”, not “up”
• High component failure rates
– Inexpensive commodity components fail all the time
• “Modest” number of huge files• “Modest” number of huge files
– Multi-gigabyte files are common, if not encouraged
• 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 (Ghemawat et al., SOSP 2003)
18.04.2012 27/39Satish Srirama
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• Single master to coordinate access, keep metadata
– Simple centralized management
• No data caching
– Little benefit due to large datasets, streaming reads
• Simplify the API
– Push some of the issues onto the client (e.g., data layout)
HDFS = GFS clone (same basic ideas implemented in Java)
18.04.2012 28/39Satish Srirama
From GFS to HDFS
• Terminology differences:
– GFS master = Hadoop namenode
– GFS chunkservers = Hadoop datanodes
• Functional differences:• Functional differences:
– No file appends in HDFS
– HDFS performance is (likely) slower
For the most part, we’ll use the Hadoop terminology…
18.04.2012 29/39Satish Srirama
(file name, block id)
(block id, block location)
HDFS namenode
File namespace/foo/bar
block 3df2
Application
HDFS Client
HDFS Architecture
Adapted from (Ghemawat et al., SOSP 2003)
instructions to datanode
datanode state(block id, byte range)
block data
HDFS datanode
Linux file system
…
HDFS datanode
Linux file system
…
18.04.2012 30Satish Srirama
Namenode Responsibilities
• Managing the file system namespace:
– Holds file/directory structure, metadata, file-to-block mapping, access permissions, etc.
• Coordinating file operations:
– Directs clients to datanodes for reads and writes– Directs clients to datanodes for reads and writes
– No data is moved through the namenode
• Maintaining overall health:
– Periodic communication with the datanodes
– Block re-replication and rebalancing
– Garbage collection
18.04.2012 31/39Satish Srirama
Hadoop Processing Model
• Create or allocate a cluster
• Put data onto the file system– Data is split into blocks
– Replicated and stored in the cluster
• Run your job• Run your job– Copy Map code to the allocated nodes
• Move computation to data, not data to computation
– Gather output of Map, sort and partition on key
– Run Reduce tasks
• Results are stored in the HDFS
18.04.2012 32/39Satish Srirama
Some MapReduce Terminology w.r.t.
Hadoop
• Job – A “full program” - an execution of a
Mapper and Reducer across a data set
• Task – An execution of a Mapper or a Reducer
on a data chunkon a data chunk
– Also known as Task-In-Progress (TIP)
• Task Attempt – A particular instance of an
attempt to execute a task on a machine
18.04.2012 33/39Satish Srirama
Terminology example
• Running “Word Count” across 20 files is one
job
• 20 files to be mapped imply 20 map tasks
+ some number of reduce tasks+ some number of reduce tasks
• At least 20 map task attempts will be
performed
– more if a machine crashes or slow, etc.
18.04.2012 34/39Satish Srirama
Task Attempts
• A particular task will be attempted at least
once, possibly more times if it crashes
– If the same input causes crashes over and over,
that input will eventually be abandonedthat input will eventually be abandoned
• Multiple attempts at one task may occur in
parallel with speculative execution turned on
– Task ID from TaskInProgress is not a unique
identifier; don’t use it that way
18.04.2012 35/39Satish Srirama
MapReduce: High Level
namenode
namenode daemon
job submission node
jobtracker
MapReduce job
submitted by
client computer
Master node
18.04.2012
datanode daemon
Linux file system
…
tasktracker
slave node
datanode daemon
Linux file system
…
tasktracker
slave node
datanode daemon
Linux file system
…
tasktracker
slave node
Task instance Task instance Task instance
36/39Satish Srirama
MapReduce/GFS Summary
• Simple, but powerful programming model
• Scales to handle petabyte+ workloads
– Google: six hours and two minutes to sort 1PB (10
trillion 100-byte records) on 4,000 computerstrillion 100-byte records) on 4,000 computers
– Yahoo!: 16.25 hours to sort 1PB on 3,800 computers
• Incremental performance improvement with
more nodes
• Seamlessly handles failures, but possibly with
performance penalties
18.04.2012 37/39Satish Srirama
Course projects
• Projects are already mentioned in
http://courses.cs.ut.ee/2012/cloud/Main/Proj
ects
• Max 3 persons per group• Max 3 persons per group
• Projects should be chosen by 25th April 2012
• If you have doubts try to meet me during this
week
18.04.2012 Satish Srirama 38/39
Next Lectures
• Note: No lecture next week
– A trip arranged by EENET
• MapReduce in different domains
• Let us have a look at some algorithms• Let us have a look at some algorithms
18.04.2012 39/39Satish Srirama
References
• Hadoop wiki http://wiki.apache.org/hadoop/
• Cloudera – Hadoop training http://www.cloudera.com/developers/learn-hadoop/training/
• J. Dean and S. Ghemawat, “MapReduce: • J. Dean and S. Ghemawat, “MapReduce: Simplified Data Processing on Large Clusters”, OSDI'04: Sixth Symposium on Operating System Design and Implementation, Dec, 2004.
• Todo:– Work with SciCloud Hadoop setup and Cloudera
virtual machine
18.04.2012 40/39Satish Srirama