Welcome to Bright Cluster Manager 9.0—the latest version of our market leading software for building HPC, data science and private cloud Linux clusters. Release 9.0 takes simplified cluster management to a whole new level with:
Enhanced public cloud integration with new data management capabilities, integrated marketplace pricing/payment options and support for larger jobs in the cloud
Combining multiple CPU architectures and operating systems in the same cluster
Auto-scale hybrid cloud infrastructure
Providing enhanced workload accounting and reporting capabilities
Run customizable dashboards in Bright View
Supporting OpenStack Stein and Ceph Nautilus
Providing additional options to deploy edge compute with increased resiliency
Determining GPU usage on a per job basis over time
Designed to improve ease of use, lower administration costs, and increase standardization across your enterprise, these platform enhancements further Bright’s mission—offering a cluster management platform that reduces complexity and decreases the effort and risk when building and maintaining HPC clusters.
"The platform enhancements included in this latest update further our mission in offering a cluster management platform that reduces complexity and decreases the effort and risk when building and maintaining Linux clusters."
-Martijn de Vries, CTO
9.0 brings with it a variety of cloud enhancements, including:
Data-Set Labeling―Makes it possible to label datasets for input and output data that are sent to the cloud only once using the cmjob utility
Support for Larger Cloud Jobs - The addition of FSx for Lustre support in AWS and Azure NetApp Files (ANF) support in Azure provides a parallel storage option that scales further than single NFS storage nodes for jobs running in these environments
Integrated Utility Pricing & Payment - Amazon AWS and Microsoft Azure are now integrated with Bright such that usage of Bright in the cloud can be billed on an hourly basis and included on your monthly statement from your cloud provider. This will streamline the purchasing process and enable you to pay for Bright Cluster Manager as an integral part of your cloud service, all in one bill.
Auto-scaling Hybrid Cloud Infrastructure - Bright’s auto-scaling capability can now automatically scale HPC queues and Kubernetes applications from pools of cloud nodes and local nodes, creating a highly effective hybrid cloud environment. Users simply specify the pools each type of workload can pull from and the order in which they would like them to be used.
Tagging of resources in AWS - When you run a compute job in AWS through the HPC workload management system, Bright will tag the instances and storage that you use in the cloud so that you can determine which jobs / users / projects were responsible for which AWS costs.
In Bright Cluster Manager 9.0, we’ve made it even easier to combine different types of servers, CPU architectures, and operating systems. This is done by making it possible to carry out package install, update, and removal functions for any CPU architecture/operating system combination from the cluster’s head node
Workload account and reporting was first announced in version 8.1, enabling administrators to track resource utilization by job and user jobs running under HPC workload managers. With 9.0, we’ve extended this feature to include jobs running on Kubernetes.
We’ve also added the ability to run multiple instances of workload accounting and reporting, making it possible to isolate resource utilization by project and limiting visibility of resource usage to the owner of the project. 9.0 also has additional reporting capabilities in Bright View which lets an administrator or user easily drill-down on resource utilization for individual users, projects and/or applications. It has also become much easier to track GPU utilization for applications during the run-time of jobs, which allows cluster owners make sure that allocated GPU resources are utilized effectively.
We enhanced Bright View’s dashboard capabilities to include customizable dashboards for entities such as nodes, node categories, and software images. This way, administrators can track all nodes in a cluster, ensure the nodes are optimized for the types of jobs they run, and make adjustments as necessary.
We are excited to announce that in 9.0, Bright OpenStack now includes the Stein release of OpenStack, and the Nautilus release of Ceph. Bright OpenStack Stein provides multi-attach for RBD volumes, optional deferred asynchronous volume deletion, bandwidth-aware VM scheduling, and a new OpenStack placement service that replaces the nova-placement-api.
In 9.0, Bright Edge now provides more options for edge deployment, including the ability to deploy a single HPC workload management system across an entire geographically distributed cluster, or deploy individual workload managers for each location. 9.0 also provides additional capabilities that further enhance the product’s ability to continue service users and edge devices in the event of network failure.
Enhancements in job metrics and accounting & reporting functionality now makes it possible to determine the GPU usage on a per job basis over time. GPUs are typically costly resources that should be used efficiently while they are allocated for a job. Jobs that reserve GPUs and then not use them, or not use them efficiently, should be easily identifiable. Administrators can get reports on how efficient allocated GPUs were used by jobs broken down by user, account/project and job name.