Category Archives: hardware

How to Plan and Configure YARN

Good (and very short) article about configuring Yarn, specifically allocating CPUs and RAM for the OS, JVMs, and MapReduce



Cisco UCS Common Platform Architecture (CPA) for Big Data with Cloudera

Cisco has created two reference architectures for Cloudera’s CDH “high performance” and “high capacity” configurations. I suppose that there’s no need for a reference architecture for “light processing” or “balanced”.


Selecting the right hardware for a Hadoop cluster

I’m summarizing this article. For specifics (such as how to configure split machines across racks to better configure the network switches) see the article. None of this content is operating system or hardware vendor specific, but generally the discussions assume Linux.

Goal is to minimize data movement and process on the same machine that stores the data. Therefore each machine in the cluster needs appropriate CPU and disk. Problem is that when building the cluster the nature of the queries and the resulting bottlenecks may not yet be known. If a business is building its first Hadoop cluster, it may not yet fully understand the types of business problems that will eventually be solved by it. That’s in contrast to a business deploying it’s Nth Oracle server.

Types of bottlenecks:

  1. IO: reading from disk (or a network location)
    • indexing
    • data import/export
    • data transformation
  2. CPU: processing the Map query
    • clustering
    • text mining
    • natural language processing

Other issues, since a cluster could eventually scale to hundreds or thousands off machines

  1. Power
  2. Cooling

The Cloudera Manager can provide realtime statistics about how a currently running MapReduce job impacts the CPU, disk, and network load.

Roles of the components within a Hadoop cluster:

  1. Name Node (and Standby Name Node): coordinating data storage on the cluster
  2. Job Tracker: coordinating data processing
  3. Task Tracker
  4. Data Node

Data Node and Task Tracker

  • The vast majority of the machines in a cluster will only peform the roles of Data Node and Task Tracker, which should not be run on the same nodes as Name and Job.
  • Other components (such as HBase) should only be run on the Data Nodes if they operate on data. You want to keep data local as much as possible. HBase needs about 16 GB Heap to avoid garbarge collection timeouts. Impala will consume up to 80% of available RAM.
  • Assumed to be lower performance machines than the Name Node and Job Tracker

Name Node and Job Tracker

  • Standby Name Node should (obviously) not be on the same machine as the Name Node.
  • Name Node (and Standby Name Node) and Job Tracker should be enterprise class machines (redundant power supplies, enterprise class raid’ed disks)
  • Name Node should have RAM in proportion to number of data blocks in the cluster. 1GB RAM for every 1 million blocks in HDFS. With 100 Data Node cluster, 64 GB RAM is fine. Since the machine’s tasks will be disk intensive, you’ll want enough RAM to minimize virtual memory swapping to disk.
  • 4 – 6 TB of disk, not raid’ed (JBOD configuration)
  • 2 CPUs (at least quad code). Recommend more CPUs and/or cores as opposed to faster CPU speed, since in a large cluster the higher speed will draw more power and generate more heat, yet not scale as well as if there were simply more CPUs or better yet nodes.

Cloudera has defined four standard configurations

  1. Light Processing (I’m not sure what the use case is for this. Prototype? Sandbox?)
  2. Balanced Compute (recommeded for your 1st cluster, since it’s not likely you’ll properly identify which configuration is best suited for your use case)
  3. Storage Heavy
  4. Compute Heavy


HDFS fault tolerance

HDFS is fault tolerant. Each file is broken up into blocks, and each block must be written to more than one server. The number of servers is configurable, but three is the common configuration. Just as with RAID, this provides fault tolerance and increase retrieval performance.

When a block is read, its checksum indicates whether the block is valid or corrupted. If corrupted, and depending on the scope of the corruption, the block may be rewriten or the server may be taken out of the cluster and the blocks spread to other existing servers. If the cluster is running within an elastic cloud then either the server is healed or a new server is added.

Unlike high end SAN hardware which is architected to avoid failure, HDFS assumes that its low end equipment will fail so it has self-healing built into its operating model.

Cheap Hardware

In theory, a big data cluster uses low cost commodity hardware (2 CPUs, 6-12 drives, 32 GB RAM). By clustering many cheap machines, high performance can be achieved at a low cost, along with high reliability due to decentralization.

There is little benefit to running Hapdoop nodes in a virtualized environment (e.g VMWare), since when the node is active (batch processing) it may be pushing RAM and CPU utilization to its limits. This is in contrast to an application or database server which has idle and bursts, but generally has constant utilization at some medium level. What is of greater benefit is a cloud implementation (e.g. Amazon Elastic Cloud) in which one can scale from a few nodes to hundreds or thousands of nodes in real time as the batch cycles through its process.

Unlike a traditional n-tier architecture, Hadoop combines compute & storage on the same box. In contrast, an Oracle cluster would typically store its databases on a SAN, and application logic would reside on yet another set of application servers which probably do not utilize their inexpensive internal drives for application specific tasks.

A Hadoop cluster is linearly scalable, up to 4000 nodes and dozens of petabytes of data.

In a traditional db cluster (such as Oracle RAC), the architecture of the cluster should be designed with knowledge of the schema and volume (input and retrieval) of the data. WIth Hadoop, scalability is, at worst, linear. Using a cloud architecture, additional Hadoop nodes can be provisioned on the fly as node utilization increases or decreases.