WHITE PAPER

Reference Guide
for Deploying
and Configuring
Apache Kafka

Version: 102
 Table of Content

1. Company Overview 3
Cloudera 3
Confluent 3
2. Introduction 3
3. Apache Kafka Technology Overview 3
4. Common Use Cases for Kafka 5
5. Deploying a Kafka Cluster 5
6 Configuring a Kafka Cluster 6
6.1 Replication, Partitions, and Leaders 6
6.2 Producers and Consumers 7
6.3 Settings for Guaranteed Message Delivery 7
7 Integrating Kafka with Other Components of Hadoop 8
7.1 Using Apache Flume to Move Data from Kafka to 8
HDFS, HBase, or Solr
7.2 Using Flume to Write Data to Kafka 10
7.3 Simple In-Flight Data Processing with 10
Flume Interceptors
7.4 Kafka and Spark Streaming for Complex 10
Real-Time Stream Processing
8 Security 11
9 Summary 11
Appendix A 11

REFERENCE GUIDE FOR DEPLOYING

AND CONFIGURING APACHE KAFKA

2
 1. Company Overviews
This paper was co-written by Cloudera and Confluent

Cloudera is revolutionizing enterprise data management by offering the first unified platform
for big data, an enterprise data hub built on Apache Hadoop. Cloudera offers enterprises one
place to store, access, process, secure, and analyze all their data, empowering them to extend
the value of existing investments while enabling fundamental new ways to derive value from
their data. Clouderas open source big data platform is the most widely adopted in the world,
and Cloudera is the most prolific contributor to the open source Hadoop ecosystem. As the
leading educator of Hadoop professionals, Cloudera has trained over 27,000 individuals
worldwide. Over 1,400 partners and a seasoned professional services team help deliver
greater time to value. Finally, only Cloudera provides proactive and predictive support to
run an enterprise data hub with confidence. Leading organizations in every industry plus
top public sector organizations globally run Cloudera in production.

Confluent is founded by the team that built Kafka at LinkedIn. They went through the
process of fully instrumenting everything that happens in a company and making it
available as realtime Kafka feeds to all data systems like Hadoop, Search, Newsfeed and so
on. At LinkedIn, this platform scaled to hundreds of billions of messages per day covering
everything happening in the company. They open sourced Kafka from its very early days,
and have led it to impressive industry-wide adoption across several thousand companies.
Now Confluent is focused on building a realtime data platform solution to help other
companies get easy access to data as realtime streams.

2. Introduction
Apache Kafka is a distributed publish-subscribe messaging system that is designed to be
fast, scalable, and durable. This open source project - licensed under the Apache license - has
gained popularity within the Hadoop ecosystem, across multiple industries. Its key strength
is the ability to make high volume data available as a real-time stream for consumption in
systems with very different requirementsfrom batch systems like Hadoop, to real-time
systems that require low-latency access, to stream processing engines like Apache Spark
Streaming that transform the data streams as they arrive. Kafkas flexibility makes it ideal for
a wide variety of use cases, from replacing traditional message brokers, to collecting user
activity data, aggregating logs, operational application metrics and device instrumentation.

This reference paper provides an overview of the general best practices for deploying and
running Kafka as a component of Clouderas Enterprise Data Hub.

3. Apache Kafka Technology Overview

Kafkas strengths are:

High-Throughput & Low Latency: Even with very modest hardware, Kafka can
support hundreds of thousands of messages per second, with latencies as low as
a few milliseconds.
Scalability: A Kafka cluster can be elastically and transparently expanded without
downtime.
Durability & Reliability: Messages are persisted on disk and replicated within the
REFERENCE GUIDE FOR DEPLOYING cluster to prevent data loss.
AND CONFIGURING APACHE KAFKA
Fault-Tolerance: Immune to machine failure in the Kafka cluster.
High Concurrency: Ability to simultaneously handle a large number (thousands)
of diverse clients, simultaneously writing to and reading from Kafka.
3
 Kafka provides a high-level abstraction called a Topic. A Topic is a category or stream
name to which messages are published. Users define a new Topic for each new category of
messages. The clients that publish messages to a Topic are called Producers. The clients
that consume messages from a Topic are called Consumers. Producers and Consumers can
simultaneously write to and read from multiple Topics. Each Kafka cluster consists of one or
more servers called Brokers. Message data is replicated and persisted on the Brokers.

producer producer producer

kafka cluster

consumer consumer consumer

Each Kafka Topic is partitioned, and messages are ordered within each partition. Writes to
each partition are sequential, and this is one of the key aspects of Kafkas performance. The
messages in the partitions are each assigned a sequential id number called the offset that
uniquely identifies each message within the partition.
Anatomy of a Topic

Partition 0 0 1 2 3 4 5 6 7 8 9 10 11 12

Partition 1 0 1 2 3 4 5 6 7 8 9 Writes

Partition 2 0 1 2 3 4 5 6 7 8 9 10 11 12

Old New

The partitions of a Topic are distributed over the Brokers of the Kafka cluster with each
Broker handling data and requests for a share of the partitions. Each partition is replicated
across a configurable number of Brokers for fault tolerance.

In Kafka, the offset or position of the last read message from a Topics partition, are maintained
by the corresponding Consumer. A Consumer will advance its offset linearly as it reads
messages. However, the position is controlled by the Consumer and it can consume messages
in any order it likes. For example, a Consumer can reset to an older offset to reprocess. To
enable this flexible consumption of messages, the Kafka cluster retains all published messages
whether or not they have been consumedfor a configurable period of time.

Detailed documentation for Kafka is available here.

REFERENCE GUIDE FOR DEPLOYING

AND CONFIGURING APACHE KAFKA

4
 4. Common Use Cases for Kafka
Log Aggregation
Kafka can be used across an organization to collect logs from multiple services and make
them available in standard format to multiple consumers, including Hadoop, Apache HBase,
and Apache Solr.

Messaging
Message Brokers are used for a variety of reasons, such as to decouple data processing
from data producers, to buffer unprocessed messages, etc. Kafka provides high-throughput,
low latency, replication, and fault-tolerance - making it a good solution for large scale
message processing applications.

Customer Activity Tracking

Kafka is often used to track the activity of customers on websites or mobile apps. User
activity (pageviews, searches, clicks, or other actions users may take) is published by
application servers to central Topics with one Topic per activity type. These Topics are
available for subscription by downstream systems for monitoring usage in real-time, and for
loading into Hadoop or offline data warehousing systems for offline processing and reporting.

Activity tracking is often very high volume as many messages are generated for each
user pageview.

Operational Metrics
Kafka is often used for logging operational monitoring data. This involves aggregating
statistics from distributed applications to produce centralized feeds of operational data,
for alerting and reporting.

Stream Processing
Many users end up doing stage-wise processing of data, where data is consumed from topics
of raw data and then aggregated, enriched, or otherwise transformed into new Kafka Topics
for further consumption. For example, a processing flow for an article recommendation
system might crawl the article content from RSS feeds and publish it to an articles Topic;
further processing might help normalize or de-duplicate this content to a Topic of cleaned
article content; and a final stage might attempt to match this content to users. This creates
a graph of real-time data flow. Spark Streaming, Storm, and Samza are popular frameworks
that are used in conjunction with Kafka to implement Stream Processing pipelines.

Event Sourcing
Event sourcing is a style of application design where state changes are logged as a time-
ordered sequence of records. Kafkas support for very large stored log data makes it an
excellent backend for an application built in this style.

5. Deploying a Kafka Cluster

Cloudera provides a Kafka Custom Service Descriptor (CSD) to enable easy deployment
and administration of a Kafka cluster via Cloudera Manager. The CSD provides granular
real-time view of the health of your Kafka Brokers, along with reporting and diagnostic
tools. This CSD is available on Clouderas downloads page.

For optimal performance, it is highly recommended that production Kafka Brokers be

deployed on dedicated machines, separate from the machines on which the rest of your
REFERENCE GUIDE FOR DEPLOYING Hadoop cluster runs. Kafka relies on dedicated disk access and large pagecache for peak
AND CONFIGURING APACHE KAFKA performance, and sharing nodes with Hadoop processes may interfere with its ability to
fully leverage the pagecache.

5
 Kafka is meant to run on industry standard hardware. The machine specs for a Kafka Broker
machine will be similar to that of your Hadoop TaskTracker or DataNodes. Though there is
no minimum specification that is required, we would suggest machines that have at least:

Processor with four 2Ghz cores

Six 7200 RPM SATA drives (JBOD or RAID10).
32GB of RAM
1Gb Ethernet
The size of your Kafka cluster will depend upon the hardware specifications of the cluster
machines, and usage factors like the number of simultaneous Producers and Consumers,
data replication parameter, data retention period, etc. For example, you need sufficient
memory to buffer data for active readers and writers. Assuming readers and writers are
fairly evenly distributed across the Brokers in your cluster, you can do a back-of-the-envelope
estimate of memory needs by assuming you want to be able to buffer for at least 30 seconds
and compute your memory need as write_throughput*30. The disk throughput is important,
and can often be a performance bottleneck, and hence more disks are often better. For a
more thorough guide to cluster sizing, please refer to Appendix A.
Kafka has a dependency on Apache Zookeeper. We would recommend a dedicated
Zookeeper cluster with three or five nodes. Remember that a larger Zookeeper cluster
will have slower writes than a smaller cluster, since data has to be propagated to a larger
quorum of machines. However, a three-node cluster can only tolerate one machine failure
for uninterrupted service, whereas a five-node cluster can tolerate two machine failures.
There is no simple formula for the correct size of a Zookeeper cluster, and we ask the user
to refer to Zookeeper documentation for more information. Note that Kafka can share a
Zookeeper cluster with other applications. However, unless you completely understand
the usage patterns of the other applications, we recommend a dedicated cluster for Kafka.
Deploying and monitoring Zookeeper clusters is very easy with Cloudera Manager, and
hence adding an additional cluster has very low overhead.

6. Configuring a Kafka Cluster

The following section assumes the user is using Kafka binaries based on Apache version 0.8.2.

6.1 Replication, Partitions, and Leaders

As described in Section 3, the data written to Kafka is replicated for fault tolerance and
durability. Kafka allows users to set a separate replication factor for each Topic. The replication
factor controls how many Brokers will replicate each message that is written. If you have
a replication factor of three then up to two Brokers can fail before you will lose access
to your data. We recommend using a replication factor of at least two, so that you can
transparently bounce machines without interrupting data consumption. However, if you
have stronger durability requirements, use a replication factor of three or above.

Topics in Kafka are partitioned, and each Topic has a configurable partition count. The
partition count controls how many logs the topic will be sharded into. Producers assign
data to partitions in round-robin fashion, to spread the data belonging to a Topic among
its partitions. Each partition is of course replicated, but one replica of each partition is
selected as a Leader Partition, and all reads and writes go to this lead partition.

Here are some factors to consider while picking an appropriate partition count for a
Topic. A partition can only be read by a single Consumer (however, a Consumer can read
many partitions). Thus, if your partition count is less than your Consumer count, many
Consumers will not receive data for that Topic. Hence, we recommend a partition count
REFERENCE GUIDE FOR DEPLOYING that is higher than the maximum number of simultaneous Consumers of the data, so that
AND CONFIGURING APACHE KAFKA each Consumer receives data. Similarly, we recommend a partition count that is higher than
the number of Brokers, so that the Leader Partitions are evenly distributed among Brokers,

6
 thus distributing the read/write load (Kafka performs random and even distribution
of partitions across Brokers). As a reference, many of our customers have Topics with
hundreds of partitions each. However, note that Kafka will need to allocate memory for
message buffer per partition. If there are a large number of partitions, make sure Kafka
starts with sufficient heap space (number of partitions times replica.fetch.max.bytes).

6.2 Producers and Consumers

Producers push data to Kafka Brokers, and Consumers pull data from Kafka Brokers. To
learn how to write data to and read from Kafka, please refer to the Kafka documentation
and examples that ship with the Kafka code base.

Your applications will write data to Brokers using instances of the KafkaProducer class,
which implements the Producer API. Producers send requests asynchronously and in
parallel, but always return a Future response object that returns the offset, as well as any
error that may have occurred when the request is complete. An important configuration
to consider while setting up a Producer is the value of the parameter acks, which is
the number of acknowledgments the Producer requires the Leader Partition to have
received (from the other replicas) before marking a request complete. If acks is set to 0,
the Producer will not wait to acknowledge receipt of data by the Brokers, and hence the
Producer can not know if data delivery failed and thus there can be data loss. However,
setting acks to 0 is likely to give high throughput. For settings that achieve guaranteed
data delivery, please refer to Section 6.3.

Kafka messages consist of a fixed-size header and variable length opaque byte array payload.
By treating the data as a byte array, Kafka enables you to use a custom serialization format
or existing popular serialization formats like Apache Avro, Protobuf, etc. Though there is no
maximum message size enforced by Kafka, we recommend writing messages that are no
more than 1MB in size. Most customers see optimal throughput with messages ranging from
1-10 KB in size.

Applications can read data from Kafka by using Kafkas Consumer APIs. Kafka has two
levels of Consumer APIs. The low-level, simple API maintains a connection to a single
Broker. This API is completely stateless, with the offset being passed in on every request,
allowing the user to maintain this metadata however they choose.

The high-level API hides the details of Brokers from the user and allows consuming off the
cluster of machines without concern for the underlying topology. It also maintains the state
of what has been consumed. The high-level API also provides the ability to subscribe to
topics that match a filter expression (i.e. either a whitelist or a blacklist regular expression).

To parallelize the consumption of data from a Topic, multiple Consumers can form a group
and jointly consume a single Topic. Each Consumer in the same group should be given the
same group_id. This group_id is provided in the configuration of the Consumer, and is your
way to tell the Consumer which group it belongs to. The Consumers in a group divide up
the partitions as fairly as possible, and each partition is consumed by exactly one Consumer
in a Consumer group.

6.3 Settings for Guaranteed Message Delivery

First, we need to understand the concept on an In-Sync Replica (commonly referred to
as ISR). For a Topic partition, an ISR is a follower replica that is caught-up with the Leader
Partition, and is situated on a Broker that is alive. Thus, if a leader replica is replaced by an
ISR, there will be no loss of data. However, if a non-ISR replica is made a Leader Partition,
some data loss is possible since it may not have the latest messages. Thus, to ensure
message delivery without data loss, the following settings are important:
REFERENCE GUIDE FOR DEPLOYING While configuring a Producer, set acks=-1. This ensures that a message is
AND CONFIGURING APACHE KAFKA
considered to be successfully delivered only after ALL the ISRs have acknowledged
writing the message.

7
 Set the Topic level configuration min.insync.replicas, which specifies the number of
replicas that must acknowledge a write, for the write to be considered successful. If
this minimum cannot be met, and acks=-1, then the Producer will raise an exception.
Set the Broker configuration param unclean.leader.election.enable to false. This essentially
means you are picking durability over availability since Kafka would avoid electing a
leader, and instead make the partition unavailable, if no ISR is available to become the
next leader safely.
A typical scenario would be to create a Topic with a replication factor of 3, set min.insync.
replicas to 2, and produce with request.required.acks set to -1. However, you can increase
these numbers is you have stronger durability requirements.

7. Integrating Kafka with Other Components of Hadoop

This section describes the seamless integration between Kafka and other data storage,
processing and serving layers of CDH.
Application
Servers
Kafka Cluster

PROCESS, ANALYZE, SERVE

BATCH STREAM SQL SEARCH SDK

UNIFIED SERVICES

RESOURCE MANAGEMENT SECURITY Enterprise

OPERATIONS DATA Data Warehouse
MANAGEMENT
FILESYSTEM RELATIONAL NoSQL

STORE

BATCH REAL-TIME

INTEGRATE

7.1 Using Apache Flume to Move Data from Kafka to HDFS, HBase, or Solr
Apache Flume is a very popular service for large scale ingestion of data into HDFS, HBase,
or Solr.

Flume has a Source, Channel, Sink architecture. A Flume Source will read data from the
datas original source, the Flume Channel will buffer the data, and the Flume Sink will write
the data out to HDFS/HBase/Solr. A Flume Agent is a (JVM) process that hosts Sources,
Sinks, and Channels.

Source Sink

Web Server
Channel
Agent

HDFS

Flume comes with an out-of-the-box library of Sources the are optimized for continuous
data ingestion from many common sources of data (syslog, http, log4j, etc). Thus, users can
perform data ingestion via configuration, without having to write a single line of code.

REFERENCE GUIDE FOR DEPLOYING

AND CONFIGURING APACHE KAFKA

8
 Flume can use Kafka as a Channel, i.e. Flume Sources write to Kafka, and Flume Sinks
can read data from Kafka. To persist a Kafka Topic to HDFS, setup Flume Agents running
Flumes HDFS Sinks, and configure these Sinks to read from Kafka (to write data to HBase
or Solr, you will use the corresponding Flume Sinks).

HDFS Sink

Kafka HDFS
Cluster or
Flume Agent HBase
or
Solr

Please review the Flume documentation for additional details on how to configure Flume
Agents. The following table describes parameters used to configure Kafka as a Flume
Channel; required properties are listed in bold.

Table 1: Flumes Kafka Channel Properties

Property Name Default Value Description
type Must be set to org.apache.flume.channel.kafka.
KafkaChannel.
brokerList The brokers the Kafka channel uses to
discover topic partitions, formatted as a
comma-separated list of hostname:port entries.
You do not need to specify the entire list of
brokers, but Cloudera recommends that you
specify at least two for high availability.
zookeeperConnect The URIs of the ZooKeeper cluster used by
Kafka. This can will be comma-separated list of
nodes in the ZooKeeper quorum (for example:
zk01.example.com:2181,zk02.example.com:2181,
zk03.example.com:2181).
topic flume-channel The Kafka topic the channel will use.
groupID flume The unique identifier of the Kafka consumer
group the channel uses to register with Kafka.
parseAsFlumeEvent true Set to true if a Flume source is writing to the
channel and will expect AvroDatums with the
FlumeEvent schema parseAsFlumeEvent true
(org.apache.flume.source.avro.
AvroFlumeEvent) in the channel. Set to false
if other producers are writing to the topic that
the channel is using
readSmallestOffset false If true will read all data in the topic. If false will
only read data written after the channel has
started. Only used when parseAsFlumeEvent
is false.
consumer.timeout.ms 100 kafka.consumer.timeout.ms (polling interval
when writing to the sink)
Other properties Used to configure the Kafka producer. You
supported by the can use any producer properties supported by
Kafka producer Kafka. Prepend the producer property name
REFERENCE GUIDE FOR DEPLOYING with the prefix kafka. (for example, kafka.
AND CONFIGURING APACHE KAFKA compression.codec).

9
 7.2 Using Flume to Write Data to Kafka
The Flume Kafka integration also enables Flume Sources to write directly to Kafka (by setting
up Kafka as a Flume Channel). This makes it easy to write data to Kafka from data sources
and APIs such as log files, http, log4j, syslog, thrift, etc. Flume sources can often be hooked
up simply via configuration, without the need to write any code. Thus, the Flume Kafka
integration enables the user to write data into Kafka via configuration, without the need
for writing Producer code.

7.3 Simple In-Flight Data Processing with Flume Interceptors

Flume provides a key component called the Interceptor, which is part of the Flume extensibility
model, to do simple processing of incoming events (data masking, filtering, etc), as they
pass through the system. Thus, Interceptors provide a lightweight, low latency method to
process data flowing through Kafka, before writing the data out to its final destination.

To use Flume Interceptors, use Flumes Kafka Source to consume from Kafka, and configure
Flume Agents as illustrated in this diagram:

Flume Agent

Kafka Source HDFS Sink

Kafka Cluster
File or
Memory
Channel
Interceptors
HDFS

Please review the Flume documentation for additional details on how to configure Flume.

7.4 Kafka and Spark Streaming for Complex Real-Time Stream Processing
Spark Streaming is an extension of the Apache Spark platform. It enables high-throughput,
scalable processing of continuous streams of data. Combining Kafka with Spark Streaming
enables the creation of real-time complex event processing architectures, which can process
your data in seconds (or even hundreds of milliseconds). This enables the user to get deep
insights from data in near real-time.

In Spark Streaming, the abstraction that represents a continuous stream of data is called a
DStream. Spark Streaming can be configured to consume Topics from Kafka, and create
corresponding Kafka DStreams. Each DStreams batches incoming messages into an
abstraction called the RDD, which is just an immutable collection of the incoming messages.
Each RDD is a micro-batch of the incoming messages, and the micro-batching window
is configurable.

Details on initializing Kafka DStreams are provided here, along with a sample application.
While configuring a Kafka DStream, you can specify the number of parallel consumer
threads. However, the Consumers of a DStream will run on the same Spark Driver node.
Thus, to do parallel consumption of a Kafka Topic from multiple machines, instantiate
multiple Kafka DStreams, and union their corresponding RDDs before processing.

The processed data can be sent downstream to a data serving platform like HBase or a
RDBMS. Often, the processed data is published back into Kafka from Spark Streaming
(to Topics that are different from the source Topics), from which multiple downstream
applications can consume the processed data.

REFERENCE GUIDE FOR DEPLOYING

AND CONFIGURING APACHE KAFKA

10
 8. Security
At the moment, Kafka does not provide its own authentication mechanism. However,
it integrates perfectly with secured Hadoop and can even use a secured ZooKeeper
cluster. The Kafka community is currently working on adding authentication, topic level
authorization, and data encryption functionality.

9. Summary
Apache Kafka is an open source, industry standard messaging solution. It provides
unmatched throughput, reliability, durability and flexibility. It complements and is well
integrated with other components of the Hadoop ecosystem, enabling businesses to
process large volumes of data in real-time and thus derive value from the data in real-time.
Kafka provides an unparalleled platform to implement use cases that leverage big data
with High Velocity.

Appendix A
Cluster Sizing
There are many variables that go into determining the correct hardware footprint for a
Kafka cluster. The most accurate way to model your use case is to simulate the load you
expect on your own hardware, and you can do this using the load generation tools that
ship with Kafka.

However, if we want to size our cluster without simulation, a very simple rule could be to
size the cluster based on the amount of disk-space required (which can be computed from
the estimated rate at which you get data * the required data retention period).

A slightly more sophisticated estimation can be done based on network and disk
throughput requirements.

Lets walk through the details of this estimation.

Lets say we want to plan for a use case with the following characteristics:

W - MB/sec of data that will be written

R - Replication factor

C - Number of Consumer groups (i.e. the number of readers for each write)

Mostly Kafka will be limited by the disk and network throughput so lets first describe the
disk and network requirements for the cluster.
The volume of writing that will be done is W * R (i.e. each replica will write each message).
Data will be read by replicas as part of the internal cluster replication and also by Consumers.
Since all the replicas other than the master will read each write this means a read volume of
(R-1) * W for replication. In addition each of the C Consumers will read each write, so there
will be a read volume of C*W. This gives the following:

Writes: W * R
Reads: (R + C - 1) * W
However, note that reads may actually be cached, in which case no actual disk I/O will be
done. We can model the effect of caching fairly easily. If the cluster has M MB of memory,
then a write rate of W MB/sec will allow M/(W*R) seconds of writes to be cached. So a
server with 32GB of memory taking writes at 50MB/sec will serve roughly the last 10
minutes of data from cache.
REFERENCE GUIDE FOR DEPLOYING Readers may fall out of cache for a variety of reasons a slow Consumer, or a failed server
AND CONFIGURING APACHE KAFKA that recovers and needs to catch up. An easy way to model this is to assume a number of

11
lagging readers you will budget for. To model this, lets call the number of lagging readers
L. A very pessimistic assumption would be that L = R + C -1, that is that all Consumers are
lagging all the time. A more realistic assumption might be to assume no more than two
Consumers are lagging at any given time.

Based on this, we can calculate our cluster-wide I/O requirements:

Disk Throughput (Read + Write): W * R + L * W

Network Read Throughput: (R + C -1) * W

Network Write Throughput: W * R

A single server will provide a given disk throughput as well as network throughput. For
example if we had a 1 Gigabit ethernet card with full duplex, then that would give us 125MB/
sec read and 125MB/sec write; likewise 6 7200 SATA drives might give us roughly 300 MB/
sec read + write throughput. Once we know the total requirements we have, as well as what
is provided by one machine, we can divide to get the total number of machines we need.

This will give us a machine count running at maximum capacity, assuming no overhead
for network protocols, as well as perfect balance of data and load. Since there is protocol
overhead as well as imbalance, we will want to have at least 2x this ideal capacity to ensure
we have sufficient capacity.

About Cloudera
Cloudera delivers the modern platform for data management and analytics. The
worlds leading organizations trust Cloudera to help solve their most challenging
business problems with Cloudera Enterprise, the fastest, easiest, and most secure
data platform built on Apache Hadoop. Our customers can efficiently capture, store,
process, and analyze vast amounts of data, empowering them to use advanced analytics
to drive business decisions quickly, flexibly, and at lower cost than has been possible
before. To ensure our customers are successful, we offer comprehensive support,
training, and professional services. Learn more at cloudera.com.

cloudera.com
1-888-789-1488 or 1-650-362-0488
Cloudera, Inc. 1001 Page Mill Road, Palo Alto, CA 94304, USA
2015 Cloudera, Inc. All rights reserved. Cloudera and the Cloudera logo are trademarks or registered trademarks of
Cloudera Inc. in the USA and other countries. All other trademarks are the property of their respective companies.
Information is subject to change without notice.