Cloud Infrastructure Development 

Overview

The University of Edinburgh introduced its OpenStack based research cloud infrastructure at the beginning of 2017. In the first year of operation the service has been very successful in attracting usage, to the point of being forced to increase the overprovisioning on nodes to meet demand.

We are aware of continuing interest in increased capacity from schools in CSE and also in the potential to support alternative computing modes from those in EDDIE (ie. Providing windows clusters to support rendering for ECA, and the option to run Digital Scholarship hadoop scripts migrated from UCL).

While we have sufficient staff resource to operate and maintain the infrastructure and we have been able to install  components which would provide advanced features, due to resource constraints we are unable to test and certify additional features for production.

Features:

Hadoop/spark module.

Terraform – image migration/sharing module.

Container support.

Improved GP-GPU integration.

Database as a service module.

The budget of £185 (capital) will be broken down into the following. (Very High Level).

Hypervisor

15k (roughly) on Hypervisor servers. 200 cores. 4 x 4 nodes of 2U @ roughly 15 each.

Additional VGPGUs

Can be new nodes of repurposed from Eddie where there is a surplus.

Get GPGU node prices and Options.

                Get VGPGPU prices from Nvidia

                Get spec of potential nodes to support them.

Resilient Infrastructure Management

Currently 16 in service.

Replace bringing down to 12 worker nodes.

4 – 5 k (roughly per node).

Get specs for nodes and process.

Review the addition features added to the cloud last year and Identify potential user requirements, options, feasibility, etc.

Features:

Hadoop/spark module.

Terraform – image migration/sharing module.

Container support.

Improved GP-GPU integration.

Database as a service module.

Identify potential user requirements, options, feasibility, etc.