Bright Cluster Manager

Installing Bright Cluster Manager doesn’t require Linux, HPC, Hadoop or OpenStack expertise. The installation is highly flexible with many options to customize, and almost every setting configured during installation can be changed afterwards.

• Installs directly on physical servers — no extra software, not even an operating system, required.
• Software compatibility is guaranteed.
• The installer pulls in every infrastructure component you may need — Apache Hadoop, Apache Spark and OpenStack included — on top of your chosen Linux distribution.
• In some cases it can also be installed on top of an existing Linux installation.

Distributing OS images to nodes is the core job of any cluster management software. For large and complex clusters, powerful and flexible node and software image preparation becomes critical — and Bright Cluster Manager goes well beyond the bar.

• Stand up a node or a whole cluster in minutes.
• Build, manage and use any number of node images.
• Build images on top of different Linux kernels, or different Linux, Apache Hadoop and OpenStack distributions.
• Create or modify images without breaking software compatibility.
• Assign images to individual nodes or node groups with a single command or a few clicks.
• Modify images on the HeadNode without logging into individual cluster nodes.
• Live-update nodes from a HeadNode image without rebooting or interrupting service — instead of reinstalling.
• Run RPM package commands directly on node images.
• Update images incrementally and propagate changes to nodes.
• Configure how disks on nodes are partitioned (with optional software RAID and/or LVM).
• Prevent disks or disk partitions from being overwritten.
• Run images entirely from memory, without using local disks.
• Track changes to node images with version control.
• Roll back to a previously saved image version when needed.
• Back up the HeadNode and you’ve backed up every image.
• Update BIOS images or BIOS settings on nodes automatically — without keyboard or console access.

Bright Cluster Manager ships with a GUI designed to simplify cluster management even with minimal Linux, HPC, Hadoop or OpenStack background. The intuitive GUI delivers full visibility and control of the entire system from a single interface — no need to view or run multiple GUIs per cluster.

The GUI supports multi-user and multi-cluster management. Multiple cluster administrators can have their own logins, each with fully configurable permissions. The same GUI can also manage multiple clusters of different types in different locations.

GUI components: Menu Bar · Resources · Tabs · Main Area · Event Viewer.

Bright Cluster Manager monitors, visualizes and analyzes a comprehensive set of hardware and software metrics. Almost every software and hardware metric exposed by the Linux kernel is available, plus everything reachable via hardware management interfaces such as IPMI. The Hadoop infrastructure (JVM) and services (HDFS, YARN) come with their own broad metric set.

• Cluster Metrics — metrics for the cluster as a whole, typically aggregated and/or averaged across all nodes.
• Device Metrics — provided per individual node.

Custom metrics. Beyond the defaults, you can add custom metrics easily by hooking up a script that emits a custom metric collection. Examples include readings from an application, or from a UPS, storage device, firewall, tape robot, SAN switch or KVM switch.

Visualization. Every metric can be visualized as a chart. Line color/style, fill color/style and transparency are fully configurable. Multiple charts can be displayed in the same view, new charts created via drag-and-drop, multiple metrics compared in one chart, and charts zoomed in/out. Layouts can be saved and reused.

Rack View. Shows the cluster’s physical rack layout, displaying one or two metrics per node via a color scale. If Bright Cluster Manager knows the order and size of nodes, switches and other devices, it builds the rack view accordingly. Otherwise nodes and switches appear evenly sized and alphabetically ordered. The “zoom out” feature shows metrics across many racks as a single color map — useful for spotting hot regions in the cluster, for example.

Bright Cluster Manager includes powerful GPU management and monitoring capabilities built on the functionality exposed by NVIDIA® Tesla™ GPUs. The required CUDA and OpenCL libraries are included.

Monitored GPU metrics:

• GPU temperatures
• GPU exclusivity modes
• GPU fan speeds
• System fan speeds
• PSU voltage and current
• System LED states
• GPU ECC statistics (Fermi GPUs only)

Alerts and actions. When GPU metric thresholds are crossed, Bright Cluster Manager can automatically trigger alerts and actions. Rules are fully configurable — any internal cluster management command, Linux command or shell script can be used as an action. For example, you can easily configure rules to email you and shut down a GPU node when its temperature crosses a threshold.

Bright Cluster Manager unlocks the unbounded power of the cloud while managing the complexity for you. You can stand up a brand-new cluster in the cloud or instantly add cloud-based resources to your existing cluster. The same powerful provisioning, monitoring, scheduling and management capabilities Bright provides for on-prem clusters work in the cloud — making efficient and effective use of virtual cloud resources.

With just a few mouse clicks, and without expert Linux or cloud knowledge, you can reach a public cloud such as Amazon EC2 or Microsoft Azure.