Bright Computing, a provider of platform-independent cluster management software, has released version 8.1 of the Bright product portfolio with new capabilities for cluster workload accounting, cloud bursting, OpenStack private clouds, deep learning, AMD accelerators, Kubernetes, Ceph, and a new lightweight daemon for monitoring VMs and non-Bright clustered nodes.
‘The response to our last major release, 8.0, has been tremendous,’ said Martijn de Vries, Chief Technology Officer of Bright Computing. ‘Version 8.1 adds many new features that our customers have asked for, such as better insight into cluster utilisation and performance, cloud bursting, and more flexibility with machine learning package deployment.’
The company expanded on some of the key features included in this latest release:
New workload management accounting and reporting capabilities provide information about which cluster resources were used by different groups of users within a specified period, and how effectively those resources were used. This enables cluster administrators to track usage (both general trends and usage per user) to help determine when they’ll need to add more capacity. The reports also allow cluster administrators to fine-tune usage policies to optimise resource utilisation across the organisation.
Bright Cluster Manager for Data Science has been expanded, adding Horovod – a new high-level framework designed to make distributed deep learning fast and easy to use. Another addition is support for R via an optimised R package. Other enhancements for data science include support for CUDA 9.1, NCCL2, TensorRT, CUDA enabled OpenMPI 3.0, and optimisations for NVIDIA Volta. Python 3.6 is now supported, as are even more aggressive CPU optimisations for improved performance.
Cloud bursting to OpenStack allows Bright clusters to extend into a private OpenStack cloud for additional capacity when demand increases. This feature provides the ability to create a central cluster that serves as a pool of resources that smaller clusters in an organisation can tap into when extra capacity is needed.
The new lightweight cluster management daemon (CMDaemon) exposes Bright’s monitoring and health checking APIs so that these functions can be performed on any node, including OpenStack VMs.
Bright 8.1 also supports BlueStore, the new storage backend for Ceph, providing significant performance improvements — roughly 2x for writes.
