Join us at the OpenInfra Summit in Vancouver to discover how StackHPC is revolutionising HPC with the power of OpenStack. Find out more about our talks where we present our innovative solutions and push the boundaries of what's possible.
Graphcore provide AI solutions using their own IPU (Intelligence Processing Unit) chips, and the fastest networking interfaces and devices available. These require the most innovative and optimised infrastructure around them to allow for efficient utilisation. Maximising data ingress and processing bandwidth at every point is essential to meet the data demands of AI. The flexible network architecture of the IPU-POD products also means that the scale of the processing can be adjusted according to need, splitting or joining resources to support different sizes of workload.
Traditional deployment methods allow such infrastructure to be built but it’s very time-consuming and does not allow for rapid changes of tenant or use-model, which is why Graphcore have adopted OpenStack as the reference platform for building platforms, services and entire clouds. Using infrastructure as code (Terraform, Ansible and Packer) allows for rapid reconfiguration at every level; including RDMA via SRIOV which is critical to the function of the IPU product. Graphcore and StackHPC present a demonstration and walkthrough of the Terraform configuration.
Magnum is the OpenStack service for container orchestration. Magnum has a flexible driver architecture, open API and templated abstraction model. These give OpenStack users a simple method to create and manage deployments using various container orchestration platforms. However, over time it has solely focused and specialized on Kubernetes.
Kubernetes is rapidly evolving; lately Magnum driver development and maintenance has struggled to keep pace. This has resulted in limitations that can be frustrating for users of the important service that Magnum provides.
We present a new Magnum Kubernetes driver, based on the ClusterAPI provisioning engine. The driver harnesses upstream functionality to dramatically simplify the code base and maintenance for Magnum. It also provides a fast path to new capabilities, and leverages code widely tested across many infrastructures. At the same time the Magnum API, CLI and Horizon interface provide continuity for users.
Azimuth is an open-source, application-focused platform that empowers non-technical users by allowing them to self-provision applications on LOKI, from web-based remote desktops to complex multi-node systems and Kubernetes-based applications, using a simple web interface.
Non-technical users within an organisation often have to rely on administrators to deploy and manage the applications that they need to do their job, leading to a delay in the user becoming productive and a support burden on the administrator to perform repetitive tasks.
Here we show how Azimuth utilises LOKI and open-source automation tools to allow users to deploy and manage curated applications with just a few clicks, and how the Zenith application proxy provides secure, SSO-authenticated access to these applications. We also demonstrate how applications can be optimised to take advantage of modern hardware when required, such as GPUs and accelerated networking, transparently to the user.