Etl Big Data With Hadoop

Published on June 2016 | Categories: Documents | Downloads: 51 | Comments: 0 | Views: 403
of 9
Download PDF   Embed   Report

Comments

Content


ABSTRACT
Over the last few years, organizations across public and private sectors have made a strategic decision to turn big data into
competitive advantage. The challenge of extracting value from big data is similar in many ways to the age-old problem of
distilling business intelligence from transactional data. At the heart of this challenge is the process used to extract data
from multiple sources, transform it to fit your analytical needs, and load it into a data warehouse for subsequent analysis, a
process known as “Extract, Transform & Load” (ETL). The nature of big data requires that the infrastructure for this process
can scale cost-effectively. Apache Hadoop* has emerged as the de facto standard for managing big data. This whitepaper
examines some of the platform hardware and software considerations in using Hadoop for ETL.
– We plan to publish other white papers that show how a platform based on Apache Hadoop can be extended to
support interactive queries and real-time predictive analytics. When complete, these white papers will be available
at http://hadoop.intel.com.
The ETL Bottleneck in Big Data Analytics
Big Data refers to the large amounts, at least terabytes, of poly-structured data that flows
continuously through and around organizations, including video, text, sensor logs, and trans-
actional records. The business benefits of analyzing this data can be significant. According to
a recent study by the MIT Sloan School of Management, organizations that use analytics are
twice as likely to be top performers in their industry as those that don’t.
1

Business analysts at a large company such as Intel, for example, with its global market and
complex supply chain, have long sought insight into customer demand by analyzing far-
flung data points culled from market information and business transactions. Increasingly,
the data we need is embedded in economic reports, discussion forums, news sites, social
networks, weather reports, wikis, tweets, and blogs, as well as transactions. By analyzing all
the data available, decision-makers can better assess competitive threats, anticipate changes
in customer behavior, strengthen supply chains, improve the effectiveness of marketing
campaigns, and enhance business continuity.
Extract, Transform, and Load
Big Data with Apache Hadoop*
White Paper
Big Data Analytics
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
The ETL Bottleneck
in Big Data Analytics . . . . . . . . . . . . . . . . . . . . . . 1
Apache Hadoop for Big Data . . . . . . . . . . . . . . . 2
ETL, ELT, and ETLT
with Apache Hadoop. . . . . . . . . . . . . . . . . . . . 4
Choosing the Physical Infrastructure
for ETL with Hadoop . . . . . . . . . . . . . . . . . . . . . . 6
Compute . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Storage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2
White Paper: Extract, Transform, and Load Big Data with Apache Hadoop*
Many of these benefits are not new to organizations that have mature
processes for incorporating business intelligence (BI) and analytics into
their decision-making. However, most organizations have yet to take full
advantage of new technologies for handling big data. Put simply, the
cost of the technologies needed to store and analyze large volumes of
diverse data has dropped, thanks to open source software running on
industry-standard hardware. The cost has dropped so much, in fact,
that the key strategic question is no longer what data is relevant, but
rather how to extract the most value from all the available data.
Rapidly ingesting, storing, and processing big data requires a cost-
effective infrastructure that can scale with the amount of data
and the scope of analysis. Most organizations with traditional data
platforms—typically relational database management systems
(RDBMS) coupled to enterprise data warehouses (EDW) using ETL
tools—find that their legacy infrastructure is either technically
incapable or financially impractical for storing and analyzing big data.
A traditional ETL process extracts data from multiple sources, then
cleanses, formats, and loads it into a data warehouse for analysis. When
the source data sets are large, fast, and unstructured, traditional ETL
can become the bottleneck, because it is too complex to develop, too
expensive to operate, and takes too long to execute.
By most accounts, 80 percent of the development effort in a big
data project goes into data integration and only 20 percent goes
toward data analysis. Furthermore, a traditional EDW platform can
cost upwards of USD 60K per terabyte. Analyzing one petabyte—the
amount of data Google processes in 1 hour—would cost USD 60M.
Clearly “more of the same” is not a big data strategy that any CIO
can afford. So, enter Apache Hadoop.
Apache Hadoop for Big Data
When Yahoo, Google, Facebook, and other companies extended their
services to web-scale, the amount of data they collected routinely
from user interactions online would have overwhelmed the capabilities
of traditional IT architectures. So they built their own. In the interest of
advancing the development of core infrastructure components rapidly,
they published papers and released code for many of the components
into open source. Of these components, Apache Hadoop has rapidly
emerged as the de facto standard for managing large volumes of
unstructured data.
Apache Hadoop is an open source distributed software platform for
storing and processing data. Written in Java, it runs on a cluster of
industry-standard servers configured with direct-attached storage.
Using Hadoop, you can store petabytes of data reliably on tens of
thousands of servers while scaling performance cost-effectively by
merely adding inexpensive nodes to the cluster.
Central to the scalability of Apache Hadoop is the distributed
processing framework known as MapReduce (Figure 1). MapReduce
helps programmers solve data-parallel problems for which the data
set can be sub-divided into small parts and processed independently.
MapReduce is an important advance because it allows ordinary
developers, not just those skilled in high-performance computing,
to use parallel programming constructs without worrying about the
complex details of intra-cluster communication, task monitoring, and
failure handling. MapReduce simplifies all that.
The system splits the input data-set into multiple chunks, each of
which is assigned a map task that can process the data in parallel.
Each map task reads the input as a set of (key, value) pairs and
produces a transformed set of (key, value) pairs as the output. The
framework shuffles and sorts outputs of the map tasks, sending
the intermediate (key, value) pairs to the reduce tasks, which group
them into final results. MapReduce uses JobTracker and TaskTracker
mechanisms to schedule tasks, monitor them, and restart any that fail.
The Apache Hadoop platform also includes the Hadoop Distributed
File System (HDFS), which is designed for scalability and fault-
tolerance. HDFS stores large files by dividing them into blocks (usually
64 or 128 MB) and replicating the blocks on three or more servers.
HDFS provides APIs for MapReduce applications to read and write
data in parallel. Capacity and performance can be scaled by adding
Data Nodes, and a single NameNode mechanism manages data
placement and monitors server availability. HDFS clusters in production
use today reliably hold petabytes of data on thousands of nodes.
CRM
DATA
WAREHOUSE
OLAP
ANALYSIS
TRANSFORM
LOAD
LOAD
LOAD
DATA
MINING
REPORTING
ERP
WEB SITE
TRAFFIC
EXTRACT
3
White Paper: Extract, Transform, and Load Big Data with Apache Hadoop*
In addition to MapReduce and HDFS, Apache Hadoop includes many
other components, some of which are very useful for ETL.
• Apache Flume* is a distributed system for collecting, aggregating,
and moving large amounts of data from multiple sources into HDFS
or another central data store. Enterprises typically collect log files on
application servers or other systems and archive the log files in order
to comply with regulations. Being able to ingest and analyze that
unstructured or semi-structured data in Hadoop can turn this passive
resource into a valuable asset.
• Apache Sqoop* is a tool for transferring data between Hadoop
and relational databases. You can use Sqoop to import data from a
MySQL or Oracle database into HDFS, run MapReduce on the data,
and then export the data back into an RDBMS. Sqoop automates
these processes, using MapReduce to import and export the data
in parallel with fault-tolerance.
• Apache Hive* and Apache Pig* are programming languages that
simplify development of applications employing the MapReduce
framework. HiveQL is a dialect of SQL and supports a subset of the
syntax. Although slow, Hive is being actively enhanced by the devel-
oper community to enable low-latency queries on Apache HBase* and
HDFS. Pig Latin is a procedural programming language that provides
high-level abstractions for MapReduce. You can extend it with User
Defined Functions written in Java, Python, and other languages.
• ODBC/JDBC Connectors for HBase and Hive are often proprietary
components included in distributions for Apache Hadoop software.
They provide connectivity with SQL applications by translating stan-
dard SQL queries into HiveQL commands that can be executed upon
the data in HDFS or HBase.
Figure 1. MapReduce, the programming paradigm implemented by Apache Hadoop, breaks-up a batch job into many smaller tasks for parallel processing on a dis-
tributed system. HDFS, the distributed file system stores the data reliably.
Processing: MapReduce
Hadoop Cluster
LOGICAL ARCHITECTURE PROCESS FLOW
PHYSICAL ARCHITECTURE
Task Tracker Task Tracker Task Tracker
Rack
Split 0
Split 1
Split n
Map 0
Map 1
Map n
Reduce 0
Reduce 0
Reduce 0
Shuffle and Sort
Job Tracker
Storage: HDFS
NameNode
Data Node
BLOCK
BLOCK
Data Node
BLOCK
BLOCK
Data Node
BLOCK
BLOCK
Mapper Mapper Mapper
Reducer Reducer Reducer
Master
Slave
Slave
Slave
Slave
Slave
Slave
Slave
Slave
Input
Data
Set
Hadoop Architecture
4
White Paper: Extract, Transform, and Load Big Data with Apache Hadoop*
Hadoop is a powerful platform for big data storage and processing.
However, its extensibility and novelty renew questions around data
integration, data quality, governance, security, and a host of other
issues that enterprises with mature BI processes have long taken for
granted. Despite the many challenges of integrating Hadoop into a
traditional BI environment, ETL has proven to be a frequent use case
for Hadoop in enterprises. So what explains its popularity?
ETL, ELT, and ETLT with Apache Hadoop
ETL tools move data from one place to another by performing
three functions:
• Extract data from sources such as ERP or CRM applications.
During the extract step, you may need to collect data from several
source systems and in multiple file formats, such as flat files with
delimiters (CSV) and XML files. You may also need to collect data from
legacy systems that store data in arcane formats no one else uses
anymore. This sounds easy, but can in fact be one of the main obsta-
cles in getting an ETL solution off the ground.
• Transform that data into a common format that fits other data
in the warehouse. The transform step may include multiple data
manipulations, such as moving, splitting, translating, merging, sort-
ing, pivoting, and more. For example, a customer name might be split
into first and last names, or dates might be changed to the standard
ISO format (e.g., from 07-24-13 to 2013-07-24). Often this step also
involves validating the data against data quality rules.
• Load the data into the data warehouse for analysis. This step can
be done in batch processes or row by row, more or less in real time.
In the early days, before ETL tools existed, the only way to integrate
data from different sources was to hand-code scripts in languages
such as COBOL, RPG, and PL/SQL. Antiquated though it seems, about
45 percent of all ETL work today is still done by such hand-coded
programs. Even though they are error-prone, slow to develop, and hard
to maintain, they have loyal users who seem impervious to the charms
of ETL tools, such as Oracle Warehouse Builder,* that can generate
code for databases.
Code generators also have limitations, since they work with only a
limited set of databases and are often bundled with them. In contrast,
the next generation of ETL tools includes a general-purpose engine
that performs the transformation and a set of metadata that stores
the transformation logic. Because engine-based ETL tools such as
Pentaho Kettle* and Informatica Powercenter* are independent of
source and target data stores, they are more versatile than code-
generating tools.
A traditional ETL architecture (Figure 2) accommodates multiple ETL
iterations as well as an intermediate step, performed in the “staging
area,” that gets data out of the source systems as quickly as possible.
A staging area may use a database or just plain CSV files, which makes
the process faster than inserting data into a database table. Additional
ETL iterations may be implemented to transfer data from the EDW
into data marts, which support specific analytic purposes and end-
user tools.
Much has changed in data warehousing over the past two decades.
Chiefly, databases have become vastly more powerful. RDBMS engines
now support complex transformations in SQL, including in-database
data mining, in-database data quality validation, cleansing, profiling,
statistical algorithms, hierarchical and drill-down functionality, and
much more. It has become more efficient to perform most types of
“transformation” within the RDBMS engine.
Figure 2. The traditional ETL architecture has served enterprises well for the past 20 years and many deployments still use it.
CRM
DATA
WAREHOUSE
OLAP
ANALYSIS
T L
DATA
MINING
REPORTING
ERP
WEB SITE
TRAFFIC
E T L E T L E
Staging
Area Data
Marts
CSV
DBMS
5
White Paper: Extract, Transform, and Load Big Data with Apache Hadoop*
Under these circumstances, Hadoop brings at least two major advan-
tages to traditional ETLT:
• Ingest massive amounts of data without specifying a schema
on write. A key characteristic of Hadoop is called “no schema-
on-write,” which means you do not need to pre-define the data
schema before loading data into Hadoop. This is true not only for
structured data (such as point-of-sale transactions, call detail records,
general ledger transactions, and call center transactions), but also
for unstructured data (such as user comments, doctor’s notes, insur-
ance claims descriptions, and web logs) and social media data (from
sites such as Facebook, LinkedIn, Pinterest, and Twitter). Regardless
of whether your incoming data has explicit or implicit structure, you
can rapidly load it as-is into Hadoop, where it is available for down-
stream analytic processes.
• Offload the transformation of raw data by parallel processing at
scale. Once the data is in Hadoop (on a Hadoop-compatible file system),
you can perform the traditional ETL tasks of cleansing, normalizing,
aligning, and aggregating data for your EDW by employing the massive
scalability of MapReduce.
Hadoop allows you to avoid the transformation bottleneck in your
traditional ETLT by off-loading the ingestion, transformation, and
integration of unstructured data into your data warehouse (Figure 3).
Because Hadoop enables you to embrace more data types than
ever before, it enriches your data warehouse in ways that would
otherwise be infeasible or prohibitive. Because of its scalable perfor-
mance, you can significantly accelerate the ETLT jobs. Moreover,
because data stored in Hadoop can persist over a much longer
duration, you can provide more granular, detailed data through your
EDW for high-fidelity analysis.
As a result, ELT emerged as an alternative approach in which data is
extracted from the sources, loaded into the target database, and then
transformed and integrated into the desired format. All the heavy data
processing takes place inside the target database. The advantage of
this approach is that a database system is better suited for handling
large workloads for which hundreds of millions of records need to
be integrated. RDBMS engines are also optimized for disk I/O, which
increases throughput. And, as long as the RDBMS hardware scales up,
the system performance scales with it.
But ETL is not dead yet. Traditional ETL vendors have enhanced their
tools with pushdown SQL capabilities in which transformation can take
place either in the ETL engine (for operations not supported by the
target database) or after loading inside the database. The result is the
ETLT approach, supported equally well by leading database vendors
such as Microsoft (SQL Server Integration Services) and Oracle (Oracle
Warehouse Builder).
None of these solutions is cheap or simple, and their cost and
complexity are compounded by big data. Consider eBay, which in
2011 had over 200 million items for sale, separated into 50,000
categories, and bought and sold by 100 million registered users—
all of which entailed roughly 9 petabytes of data. Google reportedly
processes over 24 petabytes of data per day. AT&T processes
19 petabytes through their networks each day, and the video
game World of Warcraft uses 1.3 petabytes of storage. All of
these figures are already out-of-date as of this writing because
online data is growing so fast.
Figure 3. Using Apache Hadoop,* organizations can ingest, process, and export massive amounts of diverse data at scale.
CRM
DATA
WAREHOUSE
OLAP
ANALYSIS
T L
Data
Science
DATA
MINING
REPORTING
ERP
WEB SITE
TRAFFIC
SOCIAL
MEDIA
SENSOR
LOGS
E T L E T L E
Flume
Sqoop
Staging
Area Data
Marts
CSV
DBMS
Sqoop
JDBC
ODBC
Sqoop
Offoad ETL with Hadoop
6
White Paper: Extract, Transform, and Load Big Data with Apache Hadoop*
Using Hadoop in this way, the organization gains an additional ability to
store and access data that they “might” need, data that may never be
loaded into the data warehouse. For example, data scientists might want
to use the large amounts of source data from social media, web logs, or
third-party stores (from curators, such as data.gov) stored on Hadoop to
enhance new analytic models that drive research and discovery. They
can store this data cost effectively in Hadoop, and retrieve it as needed
(using Hive or other analytic tools native to the platform), without
affecting the EDW environment.
Regardless of whether your enterprise takes the ETL, ELT, or ETLT
approach to data warehousing, you can reduce the operational cost
of your overall BI/DW solution by offloading common transformation
pipelines to Apache Hadoop and using MapReduce on HDFS to provide
a scalable, fault-tolerant platform for processing large amounts of
heterogeneous data.
Choosing the Physical Infrastructure
for ETL with Hadoop
The rule of thumb for Hadoop infrastructure planning has long been
to “throw more nodes at the problem.” This is a reasonable approach
when the size of your cluster matches a web-scale challenge, such as
returning search results or personalizing web pages for hundreds of
millions of online shoppers. But a typical Hadoop cluster in an enter-
prise has around 100 nodes and is supported by IT resources that are
significantly more constrained than those of Yahoo! and Facebook.
Organizations managing these clusters must adhere to the capacity
planning and performance tuning processes typical of other IT infra-
structures. Sooner or later, they need to provision, configure, and tune
their Hadoop cluster for the specific workloads their businesses run.
Workloads vary widely, so it is important to select and configure
compute, storage, network, and software infrastructure to match
specific requirements. But before considering each of these issues,
let’s examine the common misconception that all Hadoop workloads
are I/O-bound. For the past 3 years, developers at Intel have been test-
ing the performance of successive releases of Hadoop on successive
generations of Intel processor-based servers using a reference set
of test workloads. Many of these workloads, in aggregate, resemble
real-world applications, such as ETL and analytics. Based on in-depth
instrumentation of the test clusters, we noticed that I/O and CPU
utilization vary significantly across workloads and within the stages
of MapReduce in each workload.
• TeraSort (Map: CPU Bound; Reduce: I/O Bound) TeraSort transforms
data from one representation to another. Since the size of data
remains the same from input through shuffle to output, TeraSort
tends to be I/O bound. However, when we compress map output so as
to minimize disk and network I/O during the shuffle phases, TeraSort
shows very high CPU utilization and moderate disk I/O during the map
and shuffle phases, and moderate CPU utilization and heavy disk I/O
during the reduce phase.
• WordCount (CPU Bound) WordCount extracts a small amount of inter-
esting information from a large data set, which means that the map
output and the reduce output are much smaller than the job input.
As a result, the WordCount workload is mostly CPU bound (especially
during the map stage), with high CPU utilization and light disk and
network I/O.
• Nutch Indexing (Map: CPU Bound; Reduce: I/O Bound) The input to
this workload is about 2.4 million web pages generated by crawl-
ing Wikipedia. Nutch Indexing decompresses the crawl data in the
map stage, which is CPU bound, and converts intermediate results to
inverted index files in the reduce stage, which is I/O bound.
• PageRank Search (CPU Bound) The Page Rank workload is repre-
sentative of the original inspiration for Google search based on the
PageRank algorithm for link analysis. This workload spends most of
its time on iterations of several jobs, and these jobs are generally CPU
bound, with low to medium disk I/O and memory utilization.
• Bayesian Classification (I/O Bound) This workload implements the
trainer part of the Naïve Bayesian classifier, a popular algorithm for
knowledge discovery and data mining. The workload contains four
chained Hadoop jobs, which are all mostly disk I/O bound, except for
the map tasks of the first job, which also have high CPU utilization
and happen to be the most time consuming.
• K-means Clustering (CPU Bound in iteration; I/O Bound in clustering)
This workload first computes the centroid of each cluster by running
a Hadoop job iteratively until iterations converge or the maximum
number of iterations is reached. This is CPU-bound. After that, it runs
a clustering job (I/O bound) that assigns each sample to a cluster.
Even a cursory scan of various representative workloads shows
that the system resource utilization of Hadoop is more critical for
enterprises than it has been for early adopters and more complex
than many have assumed. Intel has worked closely with a number of
customers to help develop and deploy a balanced platform for various
real-world deployments of Hadoop. Our assessments and bench-
marking efforts have led us to make several recommendations when
customers consider infrastructure hardware and distributions of
Apache Hadoop software.
Compute
The Intel® Xeon® processor E5 family provides a strong foundation for
many Hadoop workloads. A number of features built into the Intel Xeon
processor E5 family are particularly well suited for Hadoop. One feature
is Intel® Integrated I/O, which reduces I/O latency by up to 32 percent
and increases I/O bandwidth by as much as 2x.
2,3
Another is Intel® Data
7
White Paper: Extract, Transform, and Load Big Data with Apache Hadoop*
Direct I/O Technology (DDIO), which allows Intel® Ethernet adapters
to communicate directly with processor cache, rather than only with
main memory. This feature delivers more I/O bandwidth and lower
latency, which is particularly beneficial when processing large data
sets. Another feature that has captured the attention of Hadoop
users is Intel® Advanced Encryption Standard New Instructions
(AES-NI), which accelerates common cryptographic functions to erase
the performance penalty typically associated with encryption and
decryption of big-data files.
In several surveys conducted by Gartner, more than 70 percent of CIOs
reported that power and cooling issues were the largest challenge they
faced in the data center.
4
Maximizing energy-efficiency can be particu-
larly important for Hadoop clusters, which grow with data volume.
With its system on a chip (SoC) design and power envelopes as low as
6 watts, the Intel® Atom™ processor offers better density and energy-
efficiency for some workloads. Because Intel Atom processors are
based on the industry-standard x86 instruction set, they are compat-
ible with all application code that runs on Intel Xeon processors. This
uniformity helps to reduce total cost of ownership by reducing soft-
ware development, porting, and system management costs.
In addition, Intel offers a power and thermal management software
product called Intel® Data Center Manager (Intel® DCM). Intel DCM uses
the instrumentation in Intel Xeon and upcoming Intel Atom processors
and integrates with existing management consoles using standard
APIs. You can use it to monitor power and thermal data in real time for
individual servers and blades, and also for racks, rows, and logical server
groupings. You can cap power using set limits or workload-related poli-
cies, configure alerts for power and thermal events, and store and
analyze data to improve capacity planning.
Memory
Sufficient system memory is essential for high-throughput of large
numbers of parallel MapReduce tasks. Hadoop typically requires 48 GB
to 96 GB of RAM per server, and 64 GB is optimal in most cases. Always
balance server memory across available memory channels to avoid
memory-bandwidth bottlenecks.
Memory errors are one of the most common causes of server
failure and data corruption, so error-correcting code (ECC) memory
is highly recommended.
5
ECC is supported in all servers based on
Intel Xeon processors and in micro-servers based on the Intel Atom
processor family.
Storage
Each server in a Hadoop cluster requires a relatively large number of
storage drives to avoid I/O bottlenecks. Two hard drives per processor
core typically deliver good results. However, a single solid-state drive
(SSD) per core can deliver higher I/O throughput, reduced latency, and
better overall cluster performance. Intel® SSD 710 Series SATA SSDs
provide significantly faster read/write performance than mechanical
hard drives, and this extra performance can be valuable for latency-
sensitive MapReduce applications. It can also accelerate jobs as
intermediate result files are shuffled between the map and the
reduce phase.
Intel tests show that replacing mechanical drives with Intel SSDs can
increase performance by as much as 80 percent.
6
You can also use
mechanical drives and SSDs together, using Intel® Cache Acceleration
Software. This tiered storage model provides some of the performance
benefits of SSDs at a lower acquisition cost.
If you use hard drives, 7,200 RPM SATA drives provide a good balance
between cost and performance. Run the drives in the Advanced Host
Controller Interface (AHCI) mode with Native Command Queuing (NCQ)
enabled to improve performance when multiple read/write requests are
invoked simultaneously. Although you can use RAID 0 to logically combine
smaller drives into a larger pool, Hadoop automatically orchestrates data
provisioning and redundancy across nodes, so RAID is not recommended.
Network
A fast network not only allows data to be imported and exported
quickly, but can also improve performance for the shuffle phase of
MapReduce applications.
A 10 Gigabit Ethernet network provides a simple, cost-effective solu-
tion. Intel tests have shown that using 10 Gigabit Ethernet rather than
1 Gigabit Ethernet in a Hadoop cluster can improve performance for
key operations by up to 4x when using conventional hard drives. The
performance improvement is even greater when using SSDs—up to 6x.
7

The greater improvement with SSDs can be attributed to faster writes
into the storage subsystem.
As a Hadoop cluster grows to include multiple server racks, you can
scale network performance by connecting each of the 10 Gigabit
Ethernet rack-level switches to a 40 Gigabit Ethernet cluster-level
switch. As requirements continue to increase, you can interconnect
multiple cluster-level switches and add an uplink to a higher-level
switching infrastructure.
White Paper: Extract, Transform, and Load Big Data with Apache Hadoop*
8
Software
Apache Hadoop is open source software that is freely available from
the apache.org source code repository. A number of early adopters and
test deployments directly download the source and build their platform
based on the Apache Hadoop distribution. Enterprises and other orga-
nizations that need a vendor-supported platform, however, look to one
of several independent software vendors (ISVs) to provide a complete
product with software, updates, and services.
The Intel® Distribution for Apache Hadoop software (Intel® Distribution)
is an enterprise-grade software platform that includes Apache Hadoop
along with other software components. The Intel Distribution has a
number of unique features.
• Built from the silicon up for performance and security. Running
the Intel Distribution on Intel® processors enables Hadoop to fully
utilize the performance and security features available in the x86
instruction set in general and the Intel Xeon processor in particu-
lar. For example, the Intel Distribution includes enhancements that
take advantage of Intel AES-NI, available in Intel Xeon processors, to
accelerate cryptographic functions, erasing the typical performance
penalty of encryption and decryption of files in HDFS.
• Automated performance tuning. This integrated mechanism
saves time while delivering a more optimized configuration. The
Intel Distribution includes a management console that provides
an intelligent mechanism for configuring the Hadoop cluster. The
Intel Active Tuner uses a genetic algorithm to try various parameters
and converge rapidly on the optimal configuration for a given
Hadoop application.
• Support for a wide range of analytic applications. These appli-
cations come from leaders in the field, most of whom are already
collaborating with Intel at the hardware layer. For example, Intel is
working closely with Dell Kitenga, SAP, SAS, Revolution Analytics,
and several other analytics ISVs to optimize the performance and
scalability of the overall solution.
The Intel Distribution is sold with enterprise-grade support, training,
and professional services at competitive prices. It is backed by a soft-
ware team that has real-world experience with Hadoop deployments
as well as deep expertise in the full platform stack of Apache Hadoop
on Intel hardware. Download a free 90-day evaluation version of the
software from http://hadoop.intel.com and contact us for help with
complex proof-of-concept deployments.
Conclusion
The newest wave of big data is generating new opportunities and
new challenges for businesses across every industry. The challenge
of data integration—incorporating data from social media and other
unstructured data into a traditional BI environment—is one of the most
urgent issues facing CIOs and IT managers. Apache Hadoop provides
a cost-effective and massively scalable platform for ingesting big
data and preparing it for analysis. Using Hadoop to offload the tradi-
tional ETL processes can reduce time to analysis by hours or even days.
Running the Hadoop cluster efficiently means selecting an optimal
infrastructure of servers, storage, networking, and software.
Intel provides software as well as hardware platform components
to help you design and deploy an efficient, high-performing Hadoop
cluster optimized for big data ETL. Take advantage of Intel reference
architectures, training, professional services, and technical support to
speed up your deployment and reduce risk.
For more information visit
hadoop.intel.com | intel.com/bigdata | intel.com/microservers
White Paper: Extract, Transform, and Load Big Data with Apache Hadoop*
1
Source: MIT Sloan study; results published in MIT Sloan Management Review. “Big Data, Analytics and the Path From Insights to Value,” Steve LaValle, Eric Lesser, Rebecca Shockley, Michael S. Hopkins, Nina Kruschwitz, December 21, 2010.
2
Source: The claim of up to 32% reduction in I/O latency is based on Intel internal measurements of the average time for an I/O device read to local system memory under idle conditions for the Intel
®
Xeon
®
processor E5-2600 product family
versus the Intel
®
Xeon
®
processor 5600 series.
3
8 GT/s and 128b/130b encoding in PCIe 3.0 specifcation enable double the interconnect bandwidth over the PCIe 2.0 specifcation. Source: http:www.pcisig.com/news_room/November_18_2010_Press_Release/.
4
Source: Gartner 2013 CIO Survey. http://www.gartner.com/technology/cio/cioagenda.jsp
5
A multi-year study by researchers from Google and the University of Toronto verifed the importance of ECC memory. According to the team’s report “memory errors are one of the most common hardware problems to lead to machine crashes.”
They found that DRAM errors occurred in about a third of systems and more than 8 percent of DIMMs. “The conclusion we draw is that error correcting codes are critical for reducing the large number of memory errors to a manageable number
of uncorrectable errors.” Source: “DRAM Errors in the Wild: A Large-Scale Field Study,” by Bianca Schroeder, Eduardo Pinheiro, and Wolf-Dietrich Weber. www.cs.utoronto.ca/~bianca/papers/sigmetrics09.pdf
6
TeraSort Benchmarks conducted by Intel in December 2012. Custom settings: mapred.reduce.tasks=100 and mapred.job.reuse.jvm.tasks=-1. http://hadoop.apache.org/docs/current/api/org/apache/hadoop/examples/terasort/package-summa-
ry.html For a detailed description of the tests, read the Intel white paper. http://www.intel.com/content/dam/www/public/us/en/documents/white-papers/big-data-apache-hadoop-technologies-for-results-whitepaper.pdf
7
TeraSort Benchmarks conducted by Intel in December 2012. Custom settings: mapred.reduce.tasks=100 and mapred.job.reuse.jvm.num.tasks=-1. Cluster confguration: One head node (name node, job tracker), 10 workers (data nodes, task
trackers), Cisco Nexus* 5020 10 Gigabit switch. Performance measured using Iometer* with Queue Depth 32. Baseline worker node: SuperMicro SYS-1026T-URF 1U servers with two Intel
®
Xeon
®
processors X5690 @ 3.47 GHz, 48 GB RAM,
700 GB 7200 RPM SATA hard drives, Intel
®
Ethernet Server Adapter I350-T2, Apache Hadoop* 1.0.3, Red Hat Enterprise Linux* 6.3, Oracle Java* 1.7.0_05. Baseline storage: 700 GB 7200 RPM SATA hard drives, upgraded storage: Intel
®
Solid-
State Drive 520 Series (the Intel
®
Solid-State Drive 520 Series is currently not validated for data center usage). Baseline network adapter: Intel
®
Ethernet Server Adapter I350-T2, upgraded network adapter: Intel
®
Ethernet Converged Network
Adapter X520-DA2.Upgraded software in worker node: Intel
®
Distribution for Apache Hadoop* software 2.1.1. Note: Solid-state drive performance varies by capacity. More information: http://hadoop.apache.org/docs/current/api/org/apache/
hadoop/examples/terasort/package-summary.html
Software and workloads used in performance tests may have been optimized for performance only on Intel
®
microprocessors. Performance tests, such as SYSmark* and MobileMark,* are measured using specifc computer systems,
components, software, operations, and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases,
including the performance of that product when combined with other products. For more information go to www.intel.com/performance.
Results are based on Intel internal testing, using third party benchmark test data and software. Intel does not control or audit the design or implementation of third party benchmark data, software or Web sites referenced in this document. Intel
encourages all of its customers to visit the referenced Web sites or others where similar performance benchmark data are reported and confrm whether the referenced benchmark data are accurate and refect performance of systems available
for purchase.
Intel’s compilers may or may not optimize to the same degree for non-Intel microprocessors for optimizations that are not unique to Intel microprocessors. These optimizations include SSE2, SSE3, and SSE3 instruction sets and other
optimizations. Intel does not guarantee the availability, functionality, or effectiveness of any optimization on microprocessors not manufactured by Intel.
Microprocessor-dependent optimizations in this product are intended for use with Intel microprocessors. Certain optimizations not specifc to Intel microarchitecture are reserved for Intel microprocessors.
Please refer to the applicable product User and Reference Guides for more information regarding the specifc instruction sets covered by this notice.
Notice revision #20110804
INFORMATION IN THIS DOCUMENT IS PROVIDED IN CONNECTION WITH INTEL
®
PRODUCTS. NO LICENSE, EXPRESS OR IMPLIED, BY ESTOPPEL OR OTHERWISE, TO ANY INTELLECTUAL PROPERTY RIGHTS IS GRANTED
BY THIS DOCUMENT. EXCEPT AS PROVIDED IN INTEL’S TERMS AND CONDITIONS OF SALE FOR SUCH PRODUCTS, INTEL ASSUMES NO LIABILITY WHATSOEVER, AND INTEL DISCLAIMS ANY EXPRESS OR IMPLIED WAR-
RANTY, RELATING TO SALE AND/OR USE OF INTEL PRODUCTS INCLUDING LIABILITY OR WARRANTIES RELATING TO FITNESS FOR A PARTICULAR PURPOSE, MERCHANTABILITY, OR INFRINGEMENT OF ANY PATENT,
COPYRIGHT OR OTHER INTELLECTUAL PROPERTY RIGHT. UNLESS OTHERWISE AGREED IN WRITING BY INTEL, THE INTEL PRODUCTS ARE NOT DESIGNED NOR INTENDED FOR ANY APPLICATION IN WHICH THE FAILURE
OF THE INTEL PRODUCT COULD CREATE A SITUATION WHERE PERSONAL INJURY OR DEATH MAY OCCUR.
Intel may make changes to specifcations and product descriptions at any time, without notice. Designers must not rely on the absence or characteristics of any features or instructions marked “reserved” or “undefned.” Intel reserves these
for future defnition and shall have no responsibility whatsoever for conficts or incompatibilities arising from future changes to them. The information here is subject to change without notice. Do not fnalize a design with this information. The
products described in this document may contain design defects or errors known as errata which may cause the product to deviate from published specifcations. Current characterized errata are available on request. Contact your local Intel
sales offce or your distributor to obtain the latest specifcations and before placing your product order. Copies of documents which have an order number and are referenced in this document, or other Intel
literature, may be obtained by calling 1-800-548-4725, or by visiting Intel’s Web Site www.intel.com.
Copyright © 2013 Intel Corporation. All rights reserved. Intel, the Intel logo, Intel Xeon, and Intel Atom are trademarks of Intel Corporation in the U.S. and other countries.
*Other names and brands may be claimed as the property of others.
Printed in USA 0713/DF/HBD/PDF Please Recycle 329261-001US

Sponsor Documents

Or use your account on DocShare.tips

Hide

Forgot your password?

Or register your new account on DocShare.tips

Hide

Lost your password? Please enter your email address. You will receive a link to create a new password.

Back to log-in

Close