Building Out Nodes for Personal Cluster

Its time to build the cluster. I have all the parts described for my cluster in my prior post, and today’s goal is to get the computers set and fingered as described in my cluster network design. This is a rather long post as I cover both hardware and operating system set up. At the end of this process, we will have a cluster of computers that is ready to have application software installed. I suggest building all the nodes first before starting with the operating system installation. Before you get started with these steps, I highly recommend that you start the download for the Ubuntu 18.04 LTS Server installation ISO. Note that the EGLOBAL S200 computers couldn’t use the live install ISO for some reason, but everything worked just find with the alternative (or old style) Ubuntu installer. You can download the latest version of that installer here. Select the “64-bit PC (AMD64) server install image” version. Cluster Node Read More …

Project Kickoff – Personal Compute Cluster 2019 Edition

Three years ago, I worked through a project of creating a low cost, low power computer cluster, with the primary goal of becoming more familiar with the inner works of Apache Spark. This project did accomplish that goal, but since this cluster was made up of 32-bit ARM processors and each had only 2 GB of RAM, the cluster was not too useful for getting meaningful work done. What it did excel at, however, was showing the user how to write Spark code efficiently. If you wanted to get anything done on this constrained system you had to be mindful of every inefficiency. Jump ahead to 2019, and I have decided to give it another go. This time, I want to make a cluster that is moderately useful for data analysis and machine learning, but does not break the bank either. The first step is to document my system design requirements and general goals: The cluster’s primary purpose will be Read More …

Upgrading ODROID Cluster to Ubuntu 16.04

I decided to rebuild my ODROID XU4 cluster’s OS with a fresh build using Ubuntu 16.04. The previous build used Ubuntu 14.04. But seeing as 14.04 is several years old at this point, I wanted to upgrade my cluster to the most recent major build of Ubuntu Linux.  For the most part, the process of building ou the OS was pretty much the same, but there are some key differences on the master node. This post is just a wholesale copy of my previous two posts (here and here) covering the same topic, but updated for Ubuntu 16.04. Configuring the ODROID XU4 Operating System Temporary Networking Setup When setting up the nodes initially, you will need to SSH into them to configure their settings. However, if we go straight to our network design, we will not be able to connect to any node between the master node is not yet set up as a router. So we will need to Read More …

ARM7 CPUs, Double Alignment, and Apache Spark

I haven’t posted an update to my data analysis projects in a while. Partly because my day job has been a bit busy lately, and partly because what time I do have for my recreational coding has been taken up by a problem I was experiencing with Apache Spark. I started have stability problems on my ODROID XU4 cluster. I didn’t fully understand the cause at first, thinking for the longest time it was my own code. In the end, it proved to be a bug in spark, or more specifically, an incompatibility between Spark’s memory management and the ARM71 platform of my ODROID XU4 cluster. The issue has to do with how some CPUs operate on double floating point values. These CPUs, including the 32-bit ARM71 CPU found in the ODROID XU4, requires that when the CPU operates on a double floating point value the 8 bytes of memory used to contain the value should be aligned to 8 byte Read More …

Quantcast File System 1.2 for ARM71

NOTE – This article has been updated. It now assumes you have set up the cluster with Ubuntu 16.04, and it has the latest builds of QFS v1.2.1 and Spark v2.2.0. I have been using the Quantcast File System (QFS) as my primary distributed file system on my ODROID XU4 cluster.  Due to QFS’s low memory footprint, it works well with Spark, allowing me to assign as much of the ODROID XU4’s limited 2 GB RAM footprint to the Spark executor running on a node. Recently, QFS 1.2 was released. This version brings many features and updates, many not relevant to my ODROID cluster use case. However, the most notable updates relevant to the ODROID XU4 cluster include: Correct Spark’s ability to create a hive megastore on a new QFS instance (QFS-332) Improved error reporting in the QFS/HDFS shim HDFS shim for the Hadoop 2.7.2 API, which the latest versions of Spark use. In this post, I will update the ODROID XU4 Read More …

Using Custom Hive UDFs With PySpark

Using Python to develop on Apache Spark is easy and familiar for many developers. However, due to the fact that Spark runs in a JVM, when your Python code interacts with the underlying Spark system, there can be an expensive process of data serialization and deserialization between the JVM and the Python interpreter. If you do most of your data manipulation using data frames in PySpark, you generally avoid this serialization cost because the Python code ends up being more of a high-level coordinator of the data frame operations rather than doing low-level operations on the data itself. This changes if you ever write a UDF in Python. To avoid the JVM-to-Python data serialization costs, you can use a Hive UDF written in Java. Creating a Hive UDF and then using it within PySpark can be a bit circuitous, but it does speed up your PySpark data frame flows if they are using Python UDFs. To illustrate this, I will rework the Read More …

Airline Flight Data Analysis – Part 2 – Analyzing On-Time Performance

In my last post on this topic, we loaded the Airline On-Time Performance data set collected by the United States Department of Transportation into a Parquet file to greatly improve the speed at which the data can be analyzed. Now, let’s take a first look at the data by graphing the average airline-caused flight delay by airline. This is a rather straightforward analysis, but is a good one to get started with the data set. Open a Jupyter python notebook on the cluster in the first cell indicate that we will be using MatPlotLib to do graphing: %matplotlib inline Then, in the next cell load data frames for the airline on time activity and airline meta data based on the parquet files built in the last post. Note that I am using QFS as my distributed file system. If you are using HDFS, simply update the file URLs as needed. air_data = spark.read.parquet(‘qfs://master:20000/user/michael/data/airline_data’) airlines = spark.read.parquet(‘qfs://master:20000/user/michael/data/airline_id_table’) Now we are ready to process Read More …

Airline Flight Data Analysis – Part 1 – Data Preparation

UPDATE – I have a more modern version of this post with larger data sets available here. This data analysis project is to explore what insights can be derived from the Airline On-Time Performance data set collected by the United States Department of Transportation. The data can be downloaded in month chunks from the Bureau of Transportation Statistics website. The data gets downloaded as a raw CSV file, which is something that Spark can easily load. However, if you download 10+ years of data from the Bureau of Transportation Statistics (meaning you downloaded 120+ one month CSV files from the site), that would collectively represent 30+ GB of data. For commercial scale Spark clusters, 30 GB of text data is a trivial task. However, if you are running Spark on the ODROID XU4 cluster or in local mode on your Mac laptop, 30+ GB of text data is substantial. So, before we can do any analysis of the dataset, we need to Read More …

Adding a New Node to the ODROID XU4 Cluster

I recently acquired another ODROID XU4 device (and a MicroSD card for bulk storage) to add it to my XU4 cluster. This new node brings my node count to five. Adding a new node to the cluster is relatively straight forward, but there are a lot of details. In the modern datacenter, this operation would be accomplished through a package manager, which would build the new node according to an image. However, I haven’t set up package management on my cluster so I will need to sit up the new node manually. In the original cluster configuration, I used a 5-port ethernet switch for the internal network. Given that there was an open port, no additional hardware beyond the new node is needed. However, if I were to add a sixth node (or beyond), I would need to update my ethernet switch to something such as an 8-port switch. I will also note that using the 40 mm PCB spacers I originally ordered makes Read More …

Connecting to HDFS from an computer external to the cluster

Since I have set up my ODROID XU4 cluster to work with Spark from a Jupyter web notebook, one of the little annoyances I have had is how inefficient it was to transferring data into the HDFS file system on the cluster.  It involved downloading data to my Mac laptop via a web download, transferring that data to the master node, then running hdfs dfs -put  to place the data onto the file system. While I did set up my cluster to create an NFS mount, it can be very slow when transferring large files to HDFS. So, my solution to this was to install HDFS on my Mac laptop, and configure it to use the ODROID XU4 cluster. One of the requirements for an external client (e.g., your Mac laptop) to interact with HDFS on a cluster is that the external client must be able to connect to every node in the cluster. Since the master node of the cluster was Read More …