Overview
Virtualization is becoming the foundation on which next generation data centers (DCs) and compute clouds are being built. The decoupling of hardware and software provided by virtualization enables truly dynamic management of full software stacks on top of hardware platforms. This allows administrators to smoothly add and remove hardware to accommodate the resource requirements of their organization's applications. However, scalability remains a key bottleneck in managing the rapidly-growing virtualized DCs or clouds.
The ‘ensemble’ abstraction was introduced to facilitate scalable automated management of DCs by regarding each DC as a collection of autonomous partitions. A server ensemble is a self-managed collection of server machines. From a global management perspective, it is viewed as a single entity capable of hosting large sets of virtual servers. Internally, ensemble management requires advanced capabilities of automated planning and orchestration that consider aspects of security, performance, availability, network, energy, and more. Also, being autonomous, an ensemble is capable of monitoring its dynamic indicators such as resource utilization, server health and power consumption, and regulating and adapting the placement of its hosted workloads accordingly.
At the heart of the ensemble is its management software, called the Ensemble Manager (EM). The EM enables users to perceive the system as a single entity, and provides –scalability while the management complexity remains constant. The EM is designed using a hierarchical infrastructure, where a Global Ensemble Manager (GEM) controls multiple ensembles, each of which is controlled by a Local Ensemble Manager(LEM). The IBM Haifa team is providing major contributions to the LEM component, which includes advanced optimization and automation features.
Main Contributions
- Open architecture for integration of multiple domains in placement optimization and orchestration, where each domain has its placement constraints, and configuration actions during plan orchestration.
- Advanced ensemble management, integrating performance, security, and network domains
- High-availability aware placement with performance optimization
- Development and integration of compatibility domain (providing host compatibility constraints for placement and relocations)
- Contribution to future versions of IBM Director product
