Base Command Manager / Bright Cluster Manager Release Notes
Release notes for Bright 9.1
== New Features ==
* Support for cluster on demand for VMware
* Support for Ubuntu 20.04 (to be made available with the first 9.1 update: 9.1-2)
* Improved support for cluster extension, use of Azure marketplace node images by default
* Support for creating user home directories on the cloud or edge directors
* Support for deploying OpenShift within a Bright cluster
* Edge director high availability
* Support for multiple (topology) switches, where each node can now have multiple switches attached to it
* Offloaded monitoring support
* Chargeback reports for Accounting & Reporting
* Drain, undrain and drainstatus commands are now applied for all of WLM, Ceph and Kubernetes
* Support for setting separate timezones per node, category, edge site, or cloud region
* Improved BIOS settings handling, including Redfish support
* Ansible modules collection for managing a Bright cluster configuration
== HPC Workload Managers ==
* Integration with the new versions: slurm20, openpbs20, and pbspro2020
* NVidia GPU autodetection for Slurm
* Improved job metrics per NVidia GPU (when more than one job uses different GPUs on the same node)
* GPU settings can be now specified during the WLM setup with cm-wlm-setup
* New Slurm queue parameters: DefMemPerGPU, DefCpuPerGPU, CpuBind, QOS, and ExclusiveUser
* Over 10 new LSF cluster parameters
* Support for LSF Suite 10.2 (commercial version)
* Allow to configure multiple default LSF queues
== Machine Learning ==
* Updated the machine learning frameworks, libraries, and tools to the latest upstream versions, whenever possible
* Introduced support for CUDA 10.2 and CUDA 11.0 packages
* Dropped support for Python 2.7 and Python 3.6 packages, continued support for python 3.7
* Introduced support for Ubuntu 20.04
== Jupyter ==
* Redesigned Jupyter integration: it now includes JupyterHub, JupyterLab and Jupyter Enterprise Gateway. Support for RHEL8/Centos8, SLES12, SLES15 and Ubuntu 18.04
* Introduced support for deploying multiple JupyterHub login nodes
* Introduced Jupyter extension to manage WLMs queues and jobs, as well as Kubernetes pods
* Introduced Jupyter extension to start VNC sessions in notebooks
* Introduced Jupyter extension to dynamically create new kernels from templates
* Introduced Jupyter Python and Bash kernel templates for WLMs and Kubernetes
* Introduced Jupyter R and PySpark kernel templates for Kubernetes
* Dropped support for Jupyter kernel packages (WLMs and Kubernetes) in favour of kernel templates
* Introduced Jupyter extension to spawn WLM jobs via magic commands (%%slurm_job, %%pbs_job, ...) in Python kernels
== Kubernetes, Container Engines and Registries ==
* Upgrade Kubernetes to v1.18
* Upgrade Docker to 19.03.13
* Upgrade NVIDIA Container Toolkit to 1.3.0
* Upgrade Singularity to 3.6.3
* Upgrade Harbor to 2.1.1
== Ceph ==
* Upgrade to Ceph Octopus on RHEL8/Centos8
== Auto Scaler ==
* Support for custom allocation prolog and epilog scripts per node
* Resource constraints for memory and GPUs are now also taken into account by the Auto Scaler
* Improved handling of node reboots depending on changes in the software image
* Improved integration with cmjob: the Auto Scaler now runs the cloud director accounting for the jobs held by cmjob
* Allow to limit which resource providers a tracker can use
* The workload priorities, which are used to decide which workload is served first, can now be taken from the WLM and Kubernetes
== Linux and Hardware Integration ==
* Removed intel-truescale-ofed package
* Updated list of mellanox packages
== pythoncm ==
* Switch to python Enum for all enumerations
== cmsh ==
* Improved node selections with generic fields
* New 2D table format
* Ability to dump in JSON format detailed information about cmdaemon entities
== Bright View ==
* Introduced new User Interface design
* Improved user experience for pages navigation
* Introduced dark mode
* Introduced User Interface for BIOS settings
* Page re-styling for Accounting & Reporting
* Introduced Chargeback dashboards
* Introduced Jupyter wizard
* Allow Rack View metrics configurations to be saved in userprofile
== User Portal ==
* Introduced new User Interface design
* Improved user experience navigation
* Introduced dark mode
== Other ==
* New switch overview for IB switches
* New min/max/sum over time operations to dumpmonitoringdata command
* Merge identical cores and memory system information data together in order to save memory
* Allow rsyslog to be offloaded to the directors or other nodes