Cluster Systems

 HPC Cluster Computing

ClusterVision have a 100% dedicated focus on high performance cluster computing (HPCC). We specialise in :

• High Performance Compute clusters
• Storage, BigData & Hadoop clusters
   

Compute Clusters

ClusterVision's compute clusters consist of a number of AMD or Intel processor based servers, typically connected by Gigabit Ethernet and a high-bandwidth, low-latency network interconnect such as InfiniBand.

The clusters are usually only connected to the outside users through one node called the master node. The master node runs all central services such as the queuing system, the job scheduler and a network file system. Furthermore, it acts as the login and compile node.

The master node is in control of all slave nodes and provides a single point of administration for the whole cluster. Sometimes more than one master node is used to provide failover redundancy or to balance the load of the central services in case of a large cluster

Storage Clusters

ClusterVision's storage clusters combine high-end commodity server hardware with high-quality SATA, SAS or FCAL (Fibre Channel) RAID units or NAS servers to build clusters of file servers with virtually unlimited storage capacity.

Storage clusters can be organised as a cluster of Network Attached Storage (NAS) servers, each with a separate large file system accessible from the outside network, or as Storage Area Network (SAN), whereby one or more servers access a number of RAID units through a Fibre Channel network. In both cases, the GPFS or Lustre file systems can be used to provide a high-performance parallel file system.

BigData and Hadoop

ClusterVision has experience in the design and build of database cluster systems for BigData and Hadoop style applications, and work with a number of companies offering commercial implementations and/or providing support for Hadoop, including Cloudera CDH and NetApp Open Solution.

Contact us - Individual customer references to ClusterVision’s existing Hadoop and BigData cluster installations in Europe are available on request.

BigData
BigData is a widely used collective term for application datasets whose size or scale prevents archiving, retrieval and analysis by traditional HPC storage architectures and relational database processes.

Examples include BigScience projects, such as the Large Hadron Collider, radio frequency identification, internet management and telecommunication records, retail, e-commerce and a range of military, surveillance and other security related applications.  

As one of the most rapidly growing areas of High Performance Computing, most of the leading hardware manufacturers, including many of ClusterVision’s closest Technology Partners have specific BigData orientated solutions.  

Hadoop
Hadoop is an open source computational framework designed for data intensive distributed applications. Originally introduced by the Apache Software Foundation, the term Hadoop is now often loosely used to refer to a number of data intensive applications and distributed computational processes. A wide variety of companies and organizations use Hadoop for both research and production, including high-profile commercial organisations such as Yahoo, Facebook and Amazon.

The application is divided into many small work fragments, each of which may be processed across any node in the cluster system. Hadoop implementations therefore typically require a specific distributed file system for data storage and retrieval which provides a very high aggregate bandwidth across the cluster. In addition to standard File Transfer Protocol (FTP), Hadoop file systems include HDFS (Hadoop Distributed File System), Amazon S3, and CloudStore.

A small Hadoop cluster will typically comprise a single master and multiple compute or data processing and tracking nodes. In a larger Hadoop cluster system, in addition to the data processing nodes, the distributed file system is typically managed through a dedicated server to host the file system index, together with a secondary node to generate snapshots of the memory structures in order to secure and prevent corruption of the file-system data.

ClusterVision's database clusters provide a fully redundant, turn-key database solution by combining elements from our compute and storage clusters with a parallel database and Bright Cluster Manager. Available databases include Oracle 11g and MySQL Cluster.

ClusterVision is world-wide Oracle partner.