Consistent AI: The Journey Together

Partners come together because of a common vision. In the open source world, that vision is directly translated into shared execution. With the rapid adoption of AI/ML, the Kubeflow project has made a tremendous amount of progress in the past year, and it is awesome to be recognized as Google Cloud Technology Partner of the Year in the Container category for the second year in a row.

Here at Cisco, we had articulated some of the AI/ML challenges in an earlier blog so we decided to embark on a shared journey with Kubeflow, a machine learning platform on Kubernetes.

Looking back at the last year, we are very proud of the community’s achievements, the close collaboration we have had with the Google Cloud/AI engineering and solutions team, and of our own Cisco’s Kubeflow team spanning multiple business units.

We would like to thank the Google Cloud teams and the entire Kubeflow community for their support.

We filled some key gaps inside Kubeflow. Some of our core contributions include:

- Kubebench – this is a key component that allows one to validate a Kubeflow cluster and execute performance benchmarks.
- Operators – we started with the Pytorch operator and then expanded to improve the design of other operators.
- Katib – this has been a recent focus area and we have worked closely with the community on the re-design of the APIs and contributed a completely new neural architecture search capability (autoML).
- Migration from Ksonnet to Kustomize – Kubeflow is being refactored internally to move to Kustomize and away from Ksonnet due to the shelving of the latter, and lastly.
- Initial central User Interface: The intial default initial UI (pre-pipelines).

In addition, we have worked together with the community on user surveys that directly influenced the roadmap and helped us converge on on-premise enterprise requirements.

As a result of the Kubeflow effort, customers are now able to deploy the complete data pipeline. Whether the data sources are inside the walls of the data center or at the ends of the earth, Cisco has the infrastructure and the management tools to support the data pipeline. With the UCS C480 ML with 8 GPUs, Cisco has the complete set of servers to support data extract, transform, and load (ETL), as well as the latest deep learning algorithms.

In addition with Cisco Intersight, customers are able to set up policies and deploy them directly from the cloud.