Getting started with Big Data

So, you’ve decided that you’re taking your organization down the route of Big Data, what components do you need? What are the available components that make Big Data work? Well. Let’s take a brief overview.

In terms of hardware, you’ll need lots of servers grouped into a very large cluster, with each server having its own internal disk drives. Ideally, you’d have Linux, but you might have Windows. And, of course, you could use Linux on System z if you have a mainframe.

You’re going to need a file system and that’s HDFS (Hadoop Distributed File System). Data in a Hadoop cluster gets broken down into smaller pieces that are called blocks, and these are distributed throughout the cluster. Any work on the data can then be performed on manageable pieces rather than on the whole mass of data.

Next you want a data store – and that’s HBase. HBase is an open source, non-relational, distributed database modelled after Google’s BigTable and is written in Java. It’s a column-oriented database management system (DBMS) that runs on top of HDFS. HBase applications are written in Java.

As a runtime, there’s MapReduce – a programming model for processing large data sets with a parallel, distributed algorithm on a cluster.

What about workload management, what options do you have for that? Your open source choices are ZooKeeper, Oozie, Jaql, Lucerne, HCatalog, Pig, and Hive. According to Apache, ZooKeeper is a centralized service for maintaining configuration information, naming, providing distributed synchronization, and providing group services. Similarly, according to Apache, Oozie is a workflow scheduler system to manage Hadoop jobs. Oozie Workflow jobs are Directed Acyclical Graphs (DAGs) of actions. Oozie Coordinator jobs are recurrent Oozie Workflow jobs triggered by time (frequency) and data availabilty. Oozie is integrated with the rest of the Hadoop stack supporting several types of Hadoop jobs out of the box (such as MapReduce, Streaming MapReduce, Pig, Hive, Sqoop, and Distcp) as well as system specific jobs (such as Java programs and shell scripts). Jaql is primarily a query language for JavaScript Object Notation (JSON). It allows both structured and non-traditional data to be processed. Lucerne is an information retrieval software library from Apache that was originally created in Java. HCatalog is a table and storage management service for data created using Hadoop. Pig, also from Apache is a platform for analysing large data sets. It consists of a high-level language for expressing data analysis programs, coupled with infrastructure for evaluating these programs. The structure of Pig programs allows substantial parallelization, which enables them to handle very large data sets. Finally on the list is Hive, which is a data warehouse system for Hadoop that facilitates easy data summarization, ad hoc queries, and the analysis of large datasets stored in Hadoop compatible file systems.

So what are your integration options? Apache Flume is a distributed, reliable, and available system for efficiently collecting, aggregating and moving large amounts of log data from many different sources to a centralized data store. There’s also Sqoop, which is a tool designed for efficiently transferring bulk data between Hadoop and structured datastores such as relational databases.

And finally, is there an open source advanced analytic engine? There is and it’s called R. R is a programming language and a software suite used for data analysis, statistical computing, and data visualization. It is highly extensible and has object-oriented features and strong graphical capabilities. It is well-suited for modelling and running advanced analytics.

That will pretty much get you started and on your way. You may feel that you’d like more integration products, some form of administration, or some kind of visualization and discovery product. But this is where you need to go to specific vendors. I’m expecting to be at liberty to talk more about how IBM is looking at this in future blogs.

Big data – where are we?

At first, people would enter information into their computers, then print it off if they wanted to share the data. Then we had networks and people could electronically share data – and then others could add to it. Pretty much all the data – even in the largest IMS database – had been entered by people or calculated from data entered by people.

But more recently, things have changed. Information stored on computers has come from other sources, for example card readers, CCTV cameras, traffic flow sensors, etc, etc. Almost any device can be given an IP address, connected to a network, and used as a source of data. All these ‘things’, that can and are being connected, has led to the use of the phrase: ‘the Internet of things’. Perhaps not the most precise description, but it indicates that the Internet is being used as a way of getting information from devices – rather than waiting for a human to type in the data.

The other development that we’re all familiar with is the growth in cloud computing. What that means is devices are connected to a nebulous source of storage and processing power. Mainframers, who have been around the block a few times, feel quite happy with this model of dumb terminals connected to some giant processing device that is some distance away and not necessarily visible to the users of the dumb terminals. This is what mainframe computing was like (and still is for some users!). Other computer professionals will recognize this as another version of the client/server model that was once so fashionable.

By having so many sources of data input, you have security and storage issues, but, perhaps more importantly, you have issues about what to do with the data. It’s almost like a person with OCD hoarding old newspaper that they never look at but can’t throw away. What can you do with these vast amounts of data?

The answer is Hadoop. According to the Web site at http://hadoop.apache.org/: “The Apache Hadoop software library is a framework that allows for the distributed processing of large data sets across clusters of computers using simple programming models. It is designed to scale up from single servers to thousands of machines, each offering local computation and storage. Rather than rely on hardware to deliver high-availability, the library itself is designed to detect and handle failures at the application layer, so delivering a highly-available service on top of a cluster of computers, each of which may be prone to failures.”

So which companies are experienced with Hadoop? Cloudera was probably the best known in the field up until recently. Other companies you may not have heard of are MapR and Hortonworks. Companies you will be familiar with are EMC and VMware who have spun off a company called Pivotal. And there’s Intel, and there’s IBM.

Let’s have a quick look at what’s out there. Apache Hive was developed by Facebook, but is now Open Source. Dremel (from Google) is published, but not yet available. Apache Drill is based on Dremel, but is still in the incubation stage. Cloudera’s Impala was inspired by Dremel. IBM’s offering is Big SQL. Hive is a data warehouse infrastructure built on top of Hadoop. It converts queries into MapReduce jobs. Impala’s SQL query system for Hadoop is Open Source. It uses C++ rather than Java. It doesn’t use MapReduce. Impala only works with Cloudera’s Distribution of Hadoop (CDH).

The Apache Thrift software framework, for scalable cross-language services development, combines a software stack with a code generation engine to build services that work efficiently and seamlessly between C++, Java, Python, PHP, Ruby, Erlang, Perl, Haskell, C#, Cocoa, JavaScript, Node.js, Smalltalk, OCaml, and Delphi and other languages.

IBM’s Big SQL is a currently a technology preview. It supports SQL, and JDBC and ODBC client drivers. IBM’s distribution of Hadoop is called BigInsights. Big SQL is similar to Hive and they can cross query. Point query is used for small queries rather than MapReduce. It supports more datatypes than Hive.

So, you can see that there’s lot’s to learn about Hadoop, and I’m sure we’ll be hearing a lot more about BigInsights and Big SQL. My advice is, if you’re looking for a career path, companies are going to need experienced Hadoop people – so get some!

Big SQL

Suppose you wanted to access ‘big data’ stored in HDFS or HBase. What would you do? Well, for many people, the first step is to find out what we’re talking about. So, let’s start with big data – it’s data that’s so large and complex that it’s difficult to process using standard and familiar database management tools or applications.

According to Wikipedia, there are issues around data capture, curation, storage, search, sharing, analysis, and visualization. You’re probably thinking: why not go back to using smaller and manageable data? It seems that people want access to larger and larger amounts of data because additional information can be gained from it – allowing people to “spot business trends, determine quality of research, prevent diseases, link legal citations, combat crime, and determine real-time roadway traffic conditions”.

Now that’s clear, what are HDFS and HBase? HDFS stands for Hadoop Distributed File System. It’s a distributed, scalable, and portable file system written in Java for the Hadoop framework. HDFS stores large files across multiple machines, and replicates the data across multiple hosts. HBase is an open source, non-relational, distributed database and is also written in Java. It was developed as part of Apache Software Foundation’s Apache Hadoop project and runs on top of HDFS (Hadoop Distributed File System), providing a fault-tolerant way of storing large quantities of data.

Each node in a Hadoop instance typically has a single namenode; a cluster of datanodes form the HDFS cluster. So what’s needed is some way to access that cluster. At the moment, the choices are basically Hive, Impala, and Big SQL.

Again, a search on Wikipedia informs me that “Hive supports analysis of large datasets stored in Hadoop-compatible file systems such as Amazon S3 filesystem. It provides an SQL-like language called HiveQL while maintaining full support for map/reduce. To accelerate queries, it provides indexes, including bitmap indexes. By default, Hive stores metadata in an embedded Apache Derby database, and other client/server databases like MySQL can optionally be used. Currently, there are three file formats supported in Hive, which are TEXTFILE, SEQUENCEFILE, and RCFILE”

The Cloudera Impala project allows users to query data, whether stored in HDFS or HBase – including SELECT, JOIN, and aggregate functions – in real time. Furthermore, it uses the same metadata, SQL syntax (Hive SQL), ODBC driver, and user interface (Hue Beeswax) as Apache Hive. To avoid latency, Impala circumvents MapReduce to directly access the data through a specialized distributed query engine.

When you look up information about these things, names like Apache, Cloudera, Amazon, Facebook, Google crop up, but not IBM. You might think that’s a bit strange. Wouldn’t IBM be the organization you’d expect to have experience of big data? I mean just think of those massive IMS databases. So, why haven’t I mentioned IBM? The answer is because I haven’t got to Big SQL yet.

IBM claims that Big SQL provides robust SQL support for the Hadoop ecosystem:

•  it has a scalable architecture;
• it supports SQL and data types available in SQL '92, plus it has some additional capabilities;
• it supports JDBC and ODBC client drivers;
• it has efficient handling of ‘point queries’;
• there are a wide variety of data sources and file formats for HDFS and HBase that it supports;
• And, although it isn’t open source, it does interoperate well with the open source ecosystem within Hadoop.

The really interesting thing about this is that all the information is available in one place – Big Data University (http://bigdatauniversity.com). I’m looking forward to taking the course. Big data isn’t going away any time soon.

Why is everyone talking about Hadoop?

Hadoop is an Apache project, which means it’s open source software, and it’s written in Java. What it does is support data-intensive distributed applications. It comes from work Google were doing and allows applications to use thousands of independent computers and petabytes of data.

Yahoo has been a big contributor to the project. The Yahoo Search Webmap is a Hadoop application that is used in every Yahoo search. Facebook claims to have the largest Hadoop cluster in the world. Other users include Amazon, eBay, LinkedIn, and Twitter. But now, there’s talk of IBM taking more than a passing interest.

According to IBM: “Apache Hadoop has two main subprojects:

• MapReduce – The framework that understands and assigns work to the nodes in a cluster.
• HDFS – A file system that spans all the nodes in a Hadoop cluster for data storage. It links together the file systems on many local nodes to make them into one big file system. HDFS assumes nodes will fail, so it achieves reliability by replicating data across multiple nodes.”

It goes on to say: “Hadoop changes the economics and the dynamics of large-scale computing. Its impact can be boiled down to four salient characteristics. Hadoop enables a computing solution that is:

• Scalable – New nodes can be added as needed, and added without needing to change data formats, how data is loaded, how jobs are written, or the applications on top.
• Cost effective – Hadoop brings massively parallel computing to commodity servers. The result is a sizeable decrease in the cost per terabyte of storage, which in turn makes it affordable to model all your data.
• Flexible – Hadoop is schema-less, and can absorb any type of data, structured or not, from any number of sources. Data from multiple sources can be joined and aggregated in arbitrary ways enabling deeper analyses than any one system can provide.
• Fault tolerant – When you lose a node, the system redirects work to another location of the data and continues processing without missing a beat.”

According to Alan Radding writing in IBM Systems Magazine (http://www.ibmsystemsmag.com/mainframe/trends/whatsnew/hadoop_mainframe/) IBM “is taking a federated approach to the big data challenge by blending traditional data management technologies with what it sees as complementary new technologies, like Hadoop, that address speed and flexibility, and are ideal for data exploration, discovery and unstructured analysis.”

Hadoop could run on any mainframe already running Java or Linux. Radding lists tools to make life easier like:

• SQOOP – imports data from relational databases into Hadoop.
• Hive – enables data to be queried using an SQL-like language called HiveQL.
• Apache Pig – a high-level platform for creating the MapReduce programs used with Hadoop.

There’s also ZooKeeper, which provides a centralized infrastructure and services that enable synchronization across a cluster.

Harry Battan, data serving manager for System z, suggests that 2,000 instances of Hadoop could run on Linux on the System z, which would make a fairly large Hadoop configuration.

Hadoop still needs to be certified for mainframe use, but sites with newer hybrid machines (z114 or z196) could have Hadoop today by putting it on their x86 blades, for which Hadoop is already certified, and it could then process data from DB2 on the mainframe. But you can see why customers might be looking to get it on their mainframes because it gives them a way to get more information out of the masses of data they already possess. And data analysis is often seen as the key to continuing business success for larger organizations.