A mixed workload is the capability to run multiple workloads (HPC and AI/Big Data) with different service-level agreement. An agile and flexible IT infrastructure with optimum utilization of resources is needed to address the demands of HPC and AI/ML workloads that requires efficient computing, storage and networking.
Designing and developing flexible IT infrastructure is paramount to addressing the needs of modern mixed workloads where HPC and AI are continuously converging. If the uses of mixed workload are going to be constant, the underlying IT infrastructure must be defined for varying degree of flexibility.
While running mixed workloads, it becomes imperative that there is optimal utilization of IT resources. For, example once the session for a particular workload has ended, the nodes associated must stop and allocated to a new workload to accomplish optimum utilization of available IT resources.
As a user, the last thing that you want is an interruption or stoppage of running workloads. Resilience should be the in-built highlight of an IT infrastructure tailored to run mixed workloads. Hence, there must be a suitable migration/failover arrangement when there is an unexpected interruption or failure.
The required IT configuration for running mixed workloads will always be different. The conceived IT infrastructure should be set up in such a what that hardware and software can be configured quickly for running a specific workload.
Running mixed workloads must be convenient enough from the user’s perspective. Any maintenance or incorporated changes must be easily made.
Tyrone’s IT architecture ensures that you can run mixed workloads such as GPU in containers and on bare metal, bare metal HPC workloads, big data, virtualization, educational videos, audio videos and virtual labs. The IT infrastructure has also been designed to facilitates virtual operation over the Skylus private cloud as shown in the figure below
Tyrone Cluster Manager underpins the entire IT architecture. With the help of TCM, cluster deployment and management are simplified. It makes the cluster agile, reliable and responsive to workloads need cutting across industries and applications. In addition, it makes the underlying hardware very fluid and easy for quick deployment and redeployment depending on uses. The Skylus software helps in creating a cloud environment for the entire hardware which can be then accessed over a private cloud network. TCM is helpful to manage bare metal operations while the Tyrone Container platform (TCP) facilitates the uses of containers for running HPC and other demanding workloads. The platform also ensures optimal utilization of resources by shutting down clusters nodes once the associated workload is closed after use. Another highlight of the architecture is that two workloads can be run parallelly on bare metal and containers. The container orchestration can be triggered to the bare metal infrastructure and vice- versa depending on user requirement.