43_Sameer_Kumar_Das2

International Journal of Emerging Technologies in Engineering Research (IJETER)
Volume 4, Issue 4, April (2016) www.ijeter.everscience.org
ISSN: 2454-6410 ©EverScience Publications 198
An Insight Look Into Pig and Its Implementation
Surajit Mohanty1
, Sameer Kumar Das2
, Himanshu Suman3
, Piyush Maharana4
, Raman Ratnakar5
1-2
Asst. professor, 3-5
B_Tech Scholar
1, 3-5
Department of Computer Science Engineering, DRIEMS, Cuttack, India
2
Department of Computer Science & Engineering, GATE, Berhampur, India
Abstract – Big Data is the term used now a days for referring to
the data sets that are large and complex. The applications used
for processing traditional data are inadequate. Now Apache
Hadoop is a platform that is used for processing of big data.
Hadoop has several tools that facilitate the processing of complex
data like hive, mahout, zookeeper, pig and others. Each of these
tools cover the different challenges faced by the users. We
focused on Pig which is a scripting language for exploring large
data sets. It has the ability to process terabytes of data for a few
lines of code. All parts of processing path are customizable by
the user, storing, filtering, grouping and joining. We discuss
about Pig in brief its features and entities. Then we take up a
sample problem using a dataset and generate the results. So that
we can create a contrast between Pig and Traditional data
processing applications. We explain about how we load and store
complex information (dataset) and processing different kind of
queries and running scripts in Pig.
Index Terms – Big Data, Pig, Filtering, Grouping, Joining.
1. INTRODUCTION
Apache Hadoop ecosystem consists of several parts i.e.
consisting of frameworks and tools [fig.1].
 Distributed processing framework (YARN
MapReduce V2)
 Hadoop Distributed File System(HDFS)
 Provisioning managing and monitoring of Hadoop
clusters tool (Ambari).
 Workflow tool (Oozie).
 Scripting tool (Pig).
 Machine learning tool (Mahout).
 Statistics tool(R connector).
 SQL query tool (HIVE).
 Columnar store tool (Hbase).
 Coordination tool (Zookeeper).
 Log collector (Flame).
 Data exchange(Sqoop)
[Fig.1 Apache Hadoop ecosystem]
Apache Pig is a high level platform used with Hadoop for
creation of MapReduce program. It raises the level of
abstraction for processing large database. In MapReduce one
can specify a map function followed by a reduce function. But
with Pig the data structures become richer, typically being
multivalued and nested. Such as ‘joins’ which are not possible
in MapReduce.
2. RELATED WORK
Pig is a high level scripting language that is used with Apache
Hadoop. Pig enables data workers to write complex data
transformations without knowing Java. Pig’s simple SQL-like
scripting language is called Pig Latin, and appeals to
developers already familiar with scripting languages and
SQL. [2]
It has two components:
 Pig latin is the language used to express dataflow.
 Pig latin programs are run in two environments named
 Local execution in single JVM
 Distributed execution in Hadoop cluster.
It is a scripting language for exploring large data sets. It has
the ability to process terabytes of data for a few lines of code.
All parts of processing path are customizable by the user,
storing, filtering, grouping and joining. Pig runs as a client
side application. It has two nodes i.e. local node and
MapReduce mode.

Local node
Local nodes are suitable only for small data sets. In this mode
Pig runs in a single JVM and access the local file system.
Map Reduce mode
Here queries are translated into MapReduce job end one run
on a Hadoop cluster. It is applicable for large datasets. Here
Pig checks the HADOOP_HOME environment variable for
which client to run.
3 WAYS FOR RUNNING PIG PROGRAMS
a) Script : Pig can a script file containing pig
commands
b) Grunt: By default Pig is to run this program. It is an
infective shell for running Pig commands.
c) Embedded: pig programs in java can be run using
Pig server classes, JDBC for running sql.
Comparison with database
PIG SQL
Data flow programming layer Declarable programming
layer
Here each step is a single
transformation.
All the sets of constraints
together define the output.
Schema can be defined at
runtime.
Tightly predefined schema.
Supports complex and nested
Data Structure.
Operates on single data
structure.
Pig Latin
A Pig latin program consists of a collection of statements.
These statements might be an operation or a command.
Statement usually terminate with a semicolon.
E.g.-
grouped_records= GROUP records by year;
Here, operators and commands are not case sensitive, whereas
aliases and function names are case sensitive. When a pig is
executed each statement is passed wholly. The interpreter
makes a logical plan for the program so far.
The DUMP statement is the execution trigger until this all
statement are planned and added and at last they are all
compiled into a physical plan and executed. [1]
Brief Overview
1. Relational Operator[3]:
 Loading and storing –LOAD, STORE,
DUMP.
 Filtering – FILTER, DISTINCT,
FOREACH, MAPREDUCE, STREAM.
 Grouping and joining – JOIN, COGROUP,
GROUP, CLOSE.
 Sorting – ORDER, RANK, LIMIT.
 Combining and splitting – UNION, SPLIT.
2. Diagnostic operator
DESCRIBE, EXPLAIN, ILLUSTRATE
3. Macro and OPF
REGISTER, DEFINE, IMPORT
Commands
 HDFS – cat, cd, cp, fs, ls, mkdir, mv.
 Map Reduce – kill
 Utility – clear, exec, help, history, quit(q), run, set
Pig programs always run in JVM. However one can use
other language such as Python and Java script also.
Expressions
Constant, Field, Projection, Map lockup, cost, Arithmetic,
Conditional, Comparison, Boolean, Flatter.
Types
Boolean, Numeric, Text, Binary, Temporal, Complex
Functions
 Eval: A function that takes one or more expression
and returns another expressions.
 Filter: A special type of eval function that returns a
logical Boolean result.
 Load: A function that specifies how to load data into
a relational from external storage.
 Store: A function those specifies how to save
contents of a relation to external storage.
Macros
Like in other language, Macros are also provided for
reusing pieces of code within Pig Latin itself. At runtime,
Pig expands the macro using the macro definition
exactly.

User Defined Function
Designers can write custom codes in pig like in any other
scripting language. Normally these are written in Java. Now
to use a new function, it has to be compiled and packaged into
a JAR file.
Dynamic invokers allow calling method directly from pig
script but repeated call for a large data set can result in
significant overload.
Like Hadoop, in Pig also data loading begin before the
mapper runs.so it is extremely important to split the input
such that it can be handled independently by each mapper [4].
Filtering the income data
Now with the huge amount of data flowing in. Filtering the
irrelevant data is crucial, so, as to improve efficiency.
FOREACH GENERATOR
This operator has a nested form to support more complex
processes. It can be used to remove fields or to generate new
ones.
STREAM
This operator allows transforming data in a relation using an
external program or script. It uses Pig Storage to serialize and
desterilized relations to and from the program’s standard input
and output stream.
Grouping and Joining Data
Joining data requires expensive work in Map Reduce but Pig
provides built in functions.
JOIN AND COGROUP
JOIN gives a set of tuple, while COGROUP create a nested
set of output tuples [5].
Some practices to be adhered to:-
 Parallelism
While in MapReduce mode the degree of parallelism
must match the size of the data set.
 Anonymous relation
Instead of defining a name for every relation. Pig allows
applying the diagnostic operator like DUMP or DESCRIBE to
latest defined relation.
 Parameter substitution
If a Pig script is called regularly (i.e. with different
parameters) we can pass the parameter during the runtime,
using –param and –param –file option.
 Dynamic parameters
Pig even allows the parameters passed in the form of
commands and scripts. Backtick supper fix a great feature.
With this support parameters can be defined in the similar
way in file or command line.
 Parameter substitution processing
It is a preprocessing step before the script is run. Substitution
happen in two modes:
 Data run mode
 Normal mode
 Performance consideration in JOIN
JOIN is one of the key advantages of Pig after Map Reduce. If
one of the datasets is small enough to fit into memory, a
replicated JOIN is using likely to provide better performance.
[3]
4. PORPOSED MODELLING
Getting the data into the required shape requires information
about data so that he data can be visualized. E.g. data
generated from social media sources, the user will require
general sense as customers searching for a particular set of
products and the user will have to understand what it is and
has to visualize it. Without some sort of context, visualization
tools are likely to be of less value to the user. A solution to
the challenge is having the proper domain expertise in place.
It means analyzers of the data must have a deep understanding
of where the data comes from, who will consume the data and
how will they interpret the information. Even if we can find
and analyze data quickly and put it in the proper context for
the target people who will be using the information, value of
data generated for decision-making purposes will be
compromised if the data generated is inaccurate or untimely.
This is the challenge faced by data analysts, but taking into
account the voluminous amount of information involved in
big data projects Again, data visualization will only prove to
be a valuable tool if the data quality is assured. Analysis
becomes difficult while extremely large amounts of
information or a variety of categories of information is
involved. One way of resolving the problem is to cluster data
into a higher-level view so that smaller groups of data become
visible. By grouping the data together, or “binning,” data can
be more effectively visualized.

In the following problem we used a publicly available million
song dataset which is a freely-available collection of audio
features and metadata for a million contemporary popular
music tracks. We worked upon it using Pig tool to get solution
to certain problems such as loading and storing the data,
finding the song density, filtering the data.
Solution Approach
As we are focusing on Pig tool which basically works upon
scripting.
Problem-1
Loading and storing the million song dataset.
Solution:
To work upon any dataset we must load the data
and define the location for storing the output
generated by the user scripts/queries.
Syntax for loading the data
1. var1 = LOAD ‘input location’ USING
PigStorage(‘t’) AS (data_name1:datatype,
data_name2 :datatype,……);
Syntax for storing the data
2. STORE var1 INTO ‘output location’ USING
PigStorage(‘t’);
Once we have defined the loading and storing
location of the dataset and the output respectively,
we can proceed with our queries.
E.g. suppose we have to filter the data.
SYNTAX:
1. var2 = FILTER var1 BY data_name value;
Example
2. filtered_songs = FILTER songs BY artisthotness
> 0.5;
Our aim was to find the top 50 songs with highest
sound densities. So we will load and define the
storing location first. Then
1. Filtered_songs = Filter songs by duration>0;
2. Song_density = FOREACH filtered_songs
GENERATE artist_name, title,
density(segments_start, duration);
3. density_ordered = ORDER song_density by
density DESC;
4. top_density = LIMIT density_ordered 50;
Output generated:
3
.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
S
5. RESULT ANALYSIS
The input provided amount to a million. Variable
Song_density has a million fields each having 3 data. Then
they were ordered in descending order. And the no of output
was limited to 50.
Input –Output Pie chart
1000000
50
Input Output

Data fields produced comparison
6. CONCLUSION
As Big data invades into more and more fields the data
becomes complex. It becomes incredibly different to process
data, with on hand management tools/traditional data
processing application. The challenges are memory, namely,
capturing, storing, searching, sorting, transferring, analyzing
and visualizing. The trend of big data is picking up in the
market. Hadoop with its tools resolve and tackle big data and
resolve the problems faced by users. Further Improvements
that can be made to Pig units are Adding the notion of
workspace to each text and Remove the boiler plate code
appear where there is more than one test method and also
adding standalone utility that reads test configuration and
generates a test report.
REFERENCES
[1] “Tom White”, The Definitive Guide 3rd Edition”, Reilly, 12-14,432-447,
May.2012
[2] M. R. Palankar, A. Iamnitchi, M. Ripeanu, and S. Garfinkel, “Amazon S3
for science grids: a viable solution?” in DADC ’08: Proceedings of the
2008 international workshop on Data-aware distributed computing, 2008,
pp. 55–64.
[3] M. Isard et al. Dryad: Distributed data-parallel programs from sequential
building blocks. In European Conference on Computer Systems
(EuroSys), pages59 {72, Lisbon, Portugal, March 21-23 2007.
[4] R. Pike, S. Dorward, R. Griesemer, and S. Quinlan, “Interpreting the
data: Parallel analysis with Sawzall,” Scientific Programming, vol. 13,
no. 4, 2005.
[5] J. Dean and S. Ghemawat. MapReduce: Simplified data processing on
large clusters. In Proc. OSDI, 2004.
1000000
50
50
3
53

43_Sameer_Kumar_Das2

More Related Content

What's hot (20)

Viewers also liked (12)

Similar to 43_Sameer_Kumar_Das2 (20)

43_Sameer_Kumar_Das2