Note to other teachers and users of these slides: We would be delighted if you found our

material useful for giving your own lectures. Feel free to use these slides verbatim, or to
modify them to fit your own needs. If you make use of a significant portion of these slides
in your own lecture, please include this message, or a link to our web site: http://www.mmds.org

CS246: Mining Massive Data Sets

Jure Leskovec, Stanford University
http://cs246.stanford.edu
Data contains value and knowledge
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 2
 ¡ But to extract the knowledge data
needs to be
§ Stored (systems)
§ Managed (databases)
§ And ANALYZED ß this class

Data Mining ≈ Big Data ≈

Predictive Analytics ≈
Data Science ≈ Machine Learning
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 3
 ¡ Data mining = extraction of actionable
information from (usually) very large
datasets, is the subject of extreme hype,
fear, and interest
¡ It’s not all about machine learning
¡ But some of it is

¡ Emphasis in CS246 on algorithms that scale

§ Parallelization often essential

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 4

 ¡ Descriptive methods
§ Find human-interpretable patterns that
describe the data
§ Example: Clustering

¡ Predictive methods
§ Use some variables to predict unknown
or future values of other variables
§ Example: Recommender systems

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 5

 ¡ This combines best of machine learning,
statistics, artificial intelligence, databases but
more stress on
§ Scalability (big data)
§ Algorithms Theory,
Algorithms
Machine
Learning
§ Computing architectures
§ Automation for handling Data Mining
large data
Database
systems

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 6

 ¡ We will learn to mine different types of data:
§ Data is high dimensional
§ Data is a graph
§ Data is infinite/never-ending
§ Data is labeled
¡ We will learn to use different models of
computation:
§ MapReduce
§ Streams and online algorithms
§ Single machine in-memory
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 7
 ¡ We will learn to solve real-world problems:
§ Recommender systems
§ Market Basket Analysis
§ Spam detection
§ Duplicate document detection
¡ We will learn various “tools”:
§ Linear algebra (SVD, Rec. Sys., Communities)
§ Optimization (stochastic gradient descent)
§ Dynamic programming (frequent itemsets)
§ Hashing (LSH, Bloom filters)
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 8
High dim. Graph Infinite Machine
Apps
data data data learning

Locality Filtering
PageRank, Recommen
sensitive data SVM
SimRank der systems
hashing streams

Network Web Decision Association

Clustering
Analysis advertising Trees Rules

Dimensional Duplicate
Spam Queries on Perceptron,
ity document
reduction Detection streams kNN detection

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 9

 I
data♥

How do you want that data?

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 10
 ¡ Campuswire Q&A + chat:
§ https://campuswire.com/c/GF955CA72/
§ Use Campuswire for all questions and public
communication
§ Search the feed before asking a duplicate question
§ Please tag your posts and please no one-liners
¡ For e-mailing course staff always use:
§ [email protected]
¡ We will post course announcements to
Campuswire (hence check it regularly!)
Auditors are welcome to sit-in & audit the class
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 13
 ¡ What is Campuswire?
§ Modern-day Piazza replacement, including both
Q&A forums and chat features (i.e., Piazza + Slack)
§ Walkthrough video:
https://www.youtube.com/watch?v=GKgIOdmILpg
§ Help center:
https://intercom.help/campuswireHQ/en

You should have already received the invite by email!

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 14

 ¡ Course website: http://cs246.stanford.edu
§ Lecture slides (at least 30min before the lecture)
§ Homework, solutions, readings posted on
Campuswire

¡ Class textbook: Mining of Massive Datasets by

A. Rajaraman, J. Ullman, and J. Leskovec
§ Sold by Cambridge Uni. Press but available for free
at http://mmds.org

¡ MOOC: www.youtube.com /channel/UC_Oao2FYkLAUlUVkBfze4jg/videos

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 15
 ¡ Office hours:
§ See course website http://cs246.stanford.edu for
TA office hours
§ We start Office Hours next week!

§ For SCPD students we will use Google Hangout

§ Link will be posted on Campuswire

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 16

 ¡ Spark tutorial and help session:
§ Friday, January 10, 3:00-4:20 PM, Skilling Auditorium

¡ Review of basic probability and proof

techniques
§ Friday, January 17, 3:00-4:10 PM, Skilling Auditorium

¡ Review of linear algebra:

§ Friday, January 17, 4:20-5:20 PM, Skilling Auditorium

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 17

 ¡ 4 longer homework: 40%
§ Four major assignments, involving programming, proofs,
algorithm development.
§ Assignments take lots of time (+20h). Start early!!
¡ How to submit?
§ Homework write-up:
§ Submit via Gradescope
§ Enroll to CS246 on Canvas, and you will be automatically added to
the course Gradescope
§ Homework code:
§ If the homework requires a code submission, you will find a
separate assignment for it on Gradescope, e.g., HW1 (Code)
§ You will get a penalty if you forget to submit code!
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 18
 ¡ Homework schedule:
Date (23:59 PST) Out In
01/09, Thu HW1
01/23, Thu HW2 HW1
02/06, Thu HW3 HW2
02/20, Thu HW4 HW3
03/05, Thu HW4

§ Two late periods for HWs for the quarter:

§ Late period expires on the following Monday 23:59 PST
§ Can use max 1 late period per HW
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 19
¡ Short weekly Colab notebooks: 20%
§ Colab notebooks are posted every Thursday
§ 10 in total, from 0 to 9, each worth 2%
§ Due one week later on Thursday 23:59 PST. No late days!
§ First 2 Colabs will be posted on Thu, including detailed
submission instructions to Gradescope
§ Colab 0 (Spark Tutorial) will be solved in real-time during Fri
recitation session!
§ Colabs require at most 1hr of work
§ few lines of code!
§ “Colab” is a free cloud service from Google, hosting
Jupyter notebooks with free access to GPU and TPU
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 20
 ¡ Final exam: 40%
§ Thursday, March 19, 12:15-3:15 PM
§ Location: TBD
§ Alternative Final exam on Wed, March 18, 6-9 PM

¡ Extra credit: proportional to your contribution

§ For participating in CampusWire discussions
§ Especially valuable are answers to questions posed by
other students
§ Reporting bugs in course materials

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 21

 ¡ Programming: Python or Java
¡ Basic Algorithms: CS161 is surely sufficient
¡ Probability: e.g., CS109 or Stats116
§ There will be a review session and a review doc is
linked from the class home page
¡ Linear algebra:
§ Another review doc + review session is available
¡ Multivariable calculus
¡ Database systems (SQL, relational algebra):
§ CS145 is sufficient but not necessary
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 22
 ¡ Each of the topics listed is important for a
small part of the course:
§ If you are missing an item of background, you
could consider just-in-time learning of the needed
material

¡ The exception is programming:

§ To do well in this course, you really need to be
comfortable with writing code in Python or Java

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 23

 ¡ We’ll follow the standard CS Dept. approach:
You can get help, but you MUST acknowledge
the help on the work you hand in
¡ Failure to acknowledge your sources is a
violation of the Honor Code

¡ We use MOSS to check the originality of your

code

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 24

 ¡ You can talk to others about the algorithm(s) to
be used to solve a homework problem;
§ As long as you then mention their name(s) on the
work you submit
¡ You should not use code of others or be looking
at code of others when you write your own:
§ You can talk to people but have to write your own
solution/code
§ If you fail to mention your sources, MOSS will catch
you, and you will be charged with an HC violation

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 25

 ¡ CS341: Project in Data Mining (Spring 2020)
§ Research project on Big Data
§ Groups of 3 students
§ We provide interesting data, computing resources
(Google Cloud) and mentoring
¡ My group has RA positions open:
§ See http://snap.stanford.edu/apply/

¡ In past years we used to run CS246H.

We won’t be able to run CS246H this year.
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 26
 ¡ CS246 is fast paced!
§ Requires programming maturity
§ Strong math skills
§ SCPD students tend to be rusty on math/theory
¡ Course time commitment:
§ Homeworks take +20h
§ Colab notebooks take about 1h
¡ Form study groups

¡ It’s going to be fun and hard work. J

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 27
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 29
 ¡ Large-scale computing for data mining
problems on commodity hardware
¡ Challenges:
§ How do you distribute computation?
§ How can we make it easy to write distributed
programs?
§ Machines fail:
§ One server may stay up 3 years (1,000 days)
§ If you have 1,000 servers, expect to lose 1/day
§ With 1M machines 1,000 machines fail every day!

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 30

 ¡ Issue:
Copying data over a network takes time
¡ Idea:
§ Bring computation to data
§ Store files multiple times for reliability
¡ Spark/Hadoop address these problems
§ Storage Infrastructure – File system
§ Google: GFS. Hadoop: HDFS
§ Programming model
§ MapReduce
§ Spark
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 31
 ¡ Problem:
§ If nodes fail, how to store data persistently?
¡ Answer:
§ Distributed File System
§ Provides global file namespace
¡ Typical usage pattern:
§ Huge files (100s of GB to TB)
§ Data is rarely updated in place
§ Reads and appends are common

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 32

 ¡ Chunk servers
§ File is split into contiguous chunks
§ Typically each chunk is 16-64MB
§ Each chunk replicated (usually 2x or 3x)
§ Try to keep replicas in different racks
¡ Master node
§ a.k.a. Name Node in Hadoop’s HDFS
§ Stores metadata about where files are stored
§ Might be replicated
¡ Client library for file access
§ Talks to master to find chunk servers
§ Connects directly to chunk servers to access data
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 33
 ¡ Reliable distributed file system
¡ Data kept in “chunks” spread across machines
¡ Each chunk replicated on different machines
§ Seamless recovery from disk or machine failure

C0 C1 D0 C1 C2 C5 C0 C5

C5 C2 C5 C3 D0 D1 … D0 C2
Chunk server 1 Chunk server 2 Chunk server 3 Chunk server N

Bring computation directly to the data!

Chunk servers also serve as compute servers
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 34
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 35
 ¡ MapReduce is a style of programming
designed for:
1. Easy parallel programming
2. Invisible management of hardware and software
failures
3. Easy management of very-large-scale data

¡ It has several implementations, including

Hadoop, Spark (used in this class), Flink, and
the original Google implementation just called
“MapReduce”
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 36
 3 steps of MapReduce
¡ Map:
§ Apply a user-written Map function to each input element
§ Mapper applies the Map function to a single element
§ Many mappers grouped in a Map task (the unit of parallelism)
§ The output of the Map function is a set of 0, 1, or more
key-value pairs.
¡ Group by key: Sort and shuffle
§ System sorts all the key-value pairs by key, and
outputs key-(list of values) pairs
¡ Reduce:
§ User-written Reduce function is applied to each
key-(list of values)
Outline stays the same, Map and Reduce change to fit the problem
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 37
 MAP:
Read input and
produces a set of
key-value pairs

Group by key:
Collect all pairs with
same key
(Hash merge, Shuffle,
Sort, Partition)

Reduce:
Collect all values
belonging to the
key and output

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 38

 All phases are distributed with many tasks doing the work
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 39
 key-value
pairs

Input Output

Mappers Reducers

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 40

 Example MapReduce task:
¡ We have a huge text document
¡ Count the number of times each
distinct word appears in the file
¡ Many applications of this:
§ Analyze web server logs to find popular URLs
§ Statistical machine translation:
§ Need to count number of times every 5-word sequence
occurs in a large corpus of documents

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 41

 Provided by the Provided by the
programmer programmer
MAP: Group by key: Reduce:
Read input and Collect all values
Collect all pairs
produces a set of belonging to the
with same key
key-value pairs key and output

data
The crew of the space

reads
shuttle Endeavor recently
(The, 1) (crew, 1)

read the
returned to Earth as (crew, 1) (crew, 1)
ambassadors, harbingers of (crew, 2)
a new era of space (of, 1) (space, 1)

sequential
exploration. Scientists at
(space, 1)
(the, 1) (the, 1)
NASA are saying that the (the, 3)

Sequentially
recent assembly of the (space, 1) (the, 1)
Dextre bot is the first step in (shuttle, 1)
a long-term space-based (shuttle, 1) (the, 1)
man/mache partnership.
(recently, 1)
(Endeavor, 1) (shuttle, 1)
'"The work we're doing now …

Only
-- the robotics we're doing - (recently, 1) (recently, 1)
- is what we're going to
need …………………….. …. …
Big document (key, value) (key, value) (key, value)
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 42
 map(key, value):
# key: document name; value: text of the document
for each word w in value:
emit(w, 1)

reduce(key, values):
# key: a word; value: an iterator over counts
result = 0
for each count v in values:
result += v
emit(key, result)

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 43

 MapReduce environment takes care of:
¡ Partitioning the input data
¡ Scheduling the program’s execution across a
set of machines
¡ Performing the group by key step
§ In practice this is is the bottleneck
¡ Handling machine failures
¡ Managing required inter-machine
communication

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 44

 ¡ Map worker failure
§ Map tasks completed or in-progress at
worker are reset to idle and rescheduled
§ Reduce workers are notified when map task is
rescheduled on another worker
¡ Reduce worker failure
§ Only in-progress tasks are reset to idle and the
reduce task is restarted

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 45

 ¡ MapReduce:
§ incurs substantial overheads due to data
replication, disk I/O, and serialization

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 47

 ¡ Two major limitations of MapReduce:
§ Difficulty of programming directly in MR
§ Many problems aren’t easily described as map-reduce
§ Performance bottlenecks, or batch not fitting the
use cases
§ Persistence to disk typically slower than in-memory work

¡ In short, MR doesn’t compose well for large

applications
§ Many times one needs to chain multiple map-
reduce steps

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 48

 ¡ MapReduce uses two “ranks” of tasks:
One for Map the second for Reduce
§ Data flows from the first rank to the second

¡ Data-Flow Systems generalize this in two ways:

1. Allow any number of tasks/ranks
2. Allow functions other than Map and Reduce
§ As long as data flow is in one direction only, we can
have the blocking property and allow recovery of
tasks rather than whole jobs
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 49
 ¡ Expressive computing system, not limited to
the map-reduce model

¡ Additions to MapReduce model:

§ Fast data sharing
§ Avoids saving intermediate results to disk
§ Caches data for repetitive queries (e.g. for machine learning)
§ General execution graphs (DAGs)
§ Richer functions than just map and reduce

¡ Compatible with Hadoop

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 50
 ¡ Open source software (Apache Foundation)
¡ Supports Java, Scala and Python

¡ Key construct/idea: Resilient Distributed Dataset

(RDD)

¡ Higher-level APIs: DataFrames & DataSets

§ Introduced in more recent versions of Spark
§ Different APIs for aggregate data, which allowed to
introduce SQL support

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 51

 Key concept Resilient Distributed Dataset (RDD)
§ Partitioned collection of records
§ Generalizes (key-value) pairs
¡ Spread across the cluster, Read-only
¡ Caching dataset in memory
§ Different storage levels available
§ Fallback to disk possible
¡ RDDs can be created from Hadoop, or by
transforming other RDDs (you can stack RDDs)
¡ RDDs are best suited for applications that
apply the same operation to all elements of a
dataset
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 52
 ¡ Transformations build RDDs through
deterministic operations on other RDDs:
§ Transformations include map, filter, join, union,
intersection, distinct
§ Lazy evaluation: Nothing computed until an action
requires it

¡ Actions to return value or export data

§ Actions include count, collect, reduce, save
§ Actions can be applied to RDDs; actions force
calculations and return values
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 53
 A: B:

F:
Stage 1 groupBy

C: D: E:

join = RDD

Stage 2 map filter Stage 3 = cached partition

¡ Supports general task graphs

¡ Pipelines functions where possible
¡ Cache-aware data reuse & locality
¡ Partitioning-aware to avoid shuffles
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 54
 ¡ DataFrame:
§ Unlike an RDD, data organized into named columns,
e.g. a table in a relational database.
§ Imposes a structure onto a distributed collection of
data, allowing higher-level abstraction
¡ Dataset:
§ Extension of DataFrame API which provides type-safe,
object-oriented programming interface (compile-time
error detection)

Both built on Spark SQL engine. both can be

converted back to an RDD
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 55
 ¡ Spark SQL

¡ Spark Streaming – stream processing of live

datastreams
¡ MLlib – scalable machine learning
¡ GraphX – graph manipulation
§ extends Spark RDD with Graph abstraction: a
directed multigraph with properties attached to
each vertex and edge

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 56

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 57
 ¡ Performance: Spark normally faster but with caveats
§ Spark can process data in-memory; Hadoop MapReduce
persists back to the disk after a map or reduce action
§ Spark generally outperforms MapReduce, but it often
needs lots of memory to perform well; if there are other
resource-demanding services or can’t fit in memory, Spark
degrades
§ MapReduce easily runs alongside other services with minor
performance differences, & works well with the 1-pass jobs
it was designed for

¡ Ease of use: Spark is easier to program (higher-level APIs)

¡ Data processing: Spark more general

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 58
 ¡ Suppose we have a large web corpus
¡ Look at the metadata file
§ Lines of the form: (URL, size, date, …)
¡ For each host, find the total number of bytes
§ That is, the sum of the page sizes for all URLs from
that particular host

¡ Other examples:
§ Link analysis and graph processing
§ Machine Learning algorithms

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 60

 ¡ Statistical machine translation:
§ Need to count number of times every 5-word
sequence occurs in a large corpus of documents

¡ Very easy with MapReduce:

§ Map:
§ Extract (5-word sequence, count) from document
§ Reduce:
§ Combine the counts

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 61

 ¡ Compute the natural join R(A,B) ⋈ S(B,C)
¡ R and S are each stored in files
¡ Tuples are pairs (a,b) or (b,c)

A B B C A C
a1 b1
⋈
b2 c1 a3 c1
a2
a3
b1
b2
b2 c2 = a3 c2
b3 c3 a4 c3
a4 b3
S
R

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 62

 ¡ Use a hash function h from B-values to 1...k
¡ A Map process turns:
§ Each input tuple R(a,b) into key-value pair (b,(a,R))
§ Each input tuple S(b,c) into (b,(c,S))

¡ Map processes send each key-value pair with

key b to Reduce process h(b)
§ Hadoop does this automatically; just tell it what k is.
¡ Each Reduce process matches all the pairs
(b,(a,R)) with all (b,(c,S)) and outputs (a,b,c).
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 63
 ¡ MapReduce is great for:
§ Problems that require sequential data access
§ Large batch jobs (not interactive, real-time)

¡ MapReduce is inefficient for problems where

random (or irregular) access to data required:
§ Graphs
§ Interdependent data
§ Machine learning
§ Comparisons of many pairs of items

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 64

¡ In MapReduce we quantify the cost of an
algorithm using
1. Communication cost = total I/O of all
processes
2. Elapsed communication cost = max of I/O
along any path
3. (Elapsed) computation cost analogous, but
count only running time of processes
Note that here the big-O notation is not the most useful
(adding more machines is always an option)

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 65

 ¡ For a map-reduce algorithm:
§ Communication cost = input file size + 2 ´ (sum of
the sizes of all files passed from Map processes to
Reduce processes) + the sum of the output sizes of
the Reduce processes.
§ Elapsed communication cost is the sum of the
largest input + output for any map process, plus
the same for any reduce process

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 66

 ¡ Either the I/O (communication) or processing
(computation) cost dominates
§ Ignore one or the other

¡ Total cost tells what you pay in rent from

your friendly neighborhood cloud

¡ Elapsed cost is wall-clock time using

parallelism

1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 67

 ¡ Total communication cost
= O(|R|+|S|+|R ⋈ S|)
¡ Elapsed communication cost = O(s)
§ We’re going to pick k and the number of Map
processes so that the I/O limit s is respected
§ We put a limit s on the amount of input or output
that any one process can have. s could be:
§ What fits in main memory
§ What fits on local disk
¡ With proper indexes, computation cost is
linear in the input + output size
§ So computation cost is like comm. cost
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 68
 Grab a handout
at the back of
the room
1/7/20 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 69