Scyld ClusterWare HPC

Scyld ClusterWare offers many important features and benefits. Click to expand.

• Fast Provisioning - Shorter Time to Productivity, Quick and Easy Scale Out

  In clusters managed by traditional cluster management tools an OS installation to local disk is required for each compute node. With this approach, a significant amount of time is spent on cluster deployment and updates. With Scyld ClusterWare an OS installation to disk is only required on the master node. Compute nodes are provisioned from the master node over the network with a lightweight, cluster-aware operating environment that resides entirely in memory. A bare-metal system becomes a compute node in less than a minute just by turning it on.

  With Scyld ClusterWare's fast provisioning capabilities less time is spent on deploying a cluster. Compute nodes can be replaced quickly and a cluster can be scaled out just by adding systems to a rack and powering them up.

• Single System Image - Less Time and Expertise required for Cluster Management

  The master node is the central point of control for the entire cluster. Compute nodes are provisioned stateless, meaning that configuration information is not stored on a compute node's local disk. Unlike traditional cluster management tools, Scyld ClusterWare is not based on the principle of replicating a set of configuration files across all nodes in a cluster. Scyld ClusterWare only uses one set of configuration files in one central directory on the master node for the entire cluster. Moreover, the on-demand transfer of libraries and device drivers from the master node to compute nodes prevents version inconsistencies of libraries and drivers between compute nodes, a common problem with traditional cluster architectures. Troubleshooting problems like mismatched MPI libraries, for example, requires a thorough understanding of Linux clusters and can take hours of time. The design of Scyld ClusterWare eliminates the possibility of version and configuration inconsistencies in a cluster altogether.

  With Scyld ClusterWare, transitioning from managing a single system to managing a cluster is easy. No deep cluster expertise is required. Scyld ClusterWare offloads system administrators from having to perform tedious cluster management tasks and enables scientists and engineers who are not cluster experts to manage their cluster without loosing focus of their core competency.

• One Process Space for the Entire Cluster - Complete Process Visibility and Control at all Times

  Scyld ClusterWare aggregates process information from all compute nodes and propagates process control signals from the master node to the respective compute node(s). This unified process space gives the cluster the appearance of a single SMP system. Distributed processes as well as single system processes are started on the master node and are automatically migrated to the least loaded compute nodes. Distributed applications that require the presence of remote startup services are also supported.

  The combination of a single process space and dynamic process migration means that a direct login into a Scyld ClusterWare compute node is never required. Managing processes in the cluster is as easy as managing processes on a single system.

• Directory Services - Seamless Integration with Existing Infrastructures

  Scyld ClusterWare combines cluster specific user- and host name resolution services with generic directory services. The resolution of cluster internal names is handled efficiently while the resolution of external host or user names can be passed on to global directory services such as NIS+ or LDAP. Integration of directory services with Scyld ClusterWare requires client access only from the master node.

  The combination of cluster specific and generic directory services makes it easy to integrate Scyld ClusterWare with existing infrastructures.

• Integrated Workload Management - Optimized Resource Utilization

  In almost every HPC environment, multiple users compete for computational resources. Resource managers in conjunction with schedulers arbitrate resource demand and supply. The scheduler starts jobs under consideration of configured resource allocation policies and the current load situation as reported by the resource manager. The Torque resource manager and scheduler is pre-configured with Scyld ClusterWare. For advanced scheduling needs Penguin Computing offers the scheduler Scyld TaskMaster. Taskmaster is an adaptation of the Moab scheduler from Cluster Resources Inc. that has been integrated for Scyld ClusterWare.

  Scyld ClusterWare enables you to align cluster use with business and project priorities.

• High Availability - No Single Point of Failure, Minimal Downtime

  Scyld ClusterWare supports active/active failover configurations for master nodes as well as concurrent operation of differing software environments.

  Scyld ClusterWare's High Availability feature minimizes the impact of master node failure and adds flexibility for different application environments.

  
• Standard Base - Runs on Commodity Hardware, Supports a Standard Linux Distribution

  Scyld ClusterWare is based on industry standards. Scyld ClusterWare runs on any system based on the x86-64 processor architecture. Scyld ClusterWare is layered on top of a standard Linux distribution, typically RedHat Enterprise Linux. To find out about the availability of Scyld ClusterWare on your preferred Linux distribution, please contact scyld@penguincomputing.com.

• Commercial Support - Get Help from the Experts

  Scyld ClusterWare is supported by Penguin Computing's cluster experts. Scyld ClusterWare has been developed in-house. Expertise down to the source code level is always at hand.