Speaker: Rishit Dagli 

Abstract: There has been increasing popularity in the area of ML however deploying ML models at scale and being able to efficiently and consistently manage: serving model, infrastructure management, model versions, canary deployments, batch predictions, model metadata, monitoring data drifts, model routing, and A/B testing is a difficult task. I may say doing these steps correctly is more difficult than training a model for which a lot of literature is available. Enter Kubernetes and TensorFlow Extended. This talk will start off by giving some more clarity about the ML deployment steps as well as introduce TensorFlow Extended, an open-source tool that integrates quite well with Kubernetes and makes the deployment process consistent and easy. I also plan to talk about two more ideas in this talk: using Istio for model routing and A/B testing as well as showing the audience how they could best use their offloading pre-processing to server with Kubernetes.

To top this off, this would be followed by a demo of deploying a model at scale.

5 Mins Break | Turnaround | Raffle x 2

Speaker: Kunal Kushwaha, Dinesh Majrekar, David Fogle

Abstract: Join the Civo team as they talk about the challenges of running Kubernetes Storage Drivers on Hyper-Converged hardware. For the past few years, Civo have been deploying stateful workloads on behalf of 1,000’s customers across the globe. Meet David, Dinesh and Kunal for a Q&A session where you can ask about how to deploy storage software, manage hardware upgrades and failures, and even how to carry out a data migration between storage

5 Mins Break | Turnaround | Raffle x 2

Speaker: Jason Benedicic

Abstract: With this talk, we will look at the history of data protection from pre-virtual, virtualized, and now containerized environments. Looking at the parallels of how data protection solutions have evolved throughout the cycle and supported the move from pizza box servers to virtual machines, and now to modern container deployments. Data protection is about more than just copying the bits and bytes from one place to another, application awareness has always played a leading role in the adoption of new technologies and getting stateful workloads into production. Along with this the ability to recover multiple applications or entire sites with disaster recovery options and business continuity planning has become table stakes for enterprise workloads. Come and learn how the market is addressing these challenges within the Kubernetes landscape.

5 Mins Break | Turnaround | Raffle x 2

Speaker: Krista Macomber

Abstract: Kubernetes is entering mainstream adoption. Over 50% of customers surveyed in a recent study by independent analyst firm Evaluator Group indicated that they are using Kubernetes in production, with 60% of the customers running more than five workloads (applications). Along with this adoption comes the challenge of complexity, as customers operate multiple container management platforms across multi-cluster, multi-cloud environments. Exacerbating this problem, customers are struggling with insufficient expertise to manage Kubernetes with existing staff, as well difficulty finding, building, or keeping new talent.

Challenges pertaining to managing Kubernetes environments increases the risk of data exposure as the number, variety and severity of ransomware attacks continues to rise. In this session, Evaluator Group will draw on customer engagements and primary research findings to explore why data protection is needed for Kubernetes applications and workloads, the unique data protection requirements that Kubernetes environments have, and what customers should look for in a Kubernetes data protection solution. Attendees will walk away with a checklist of critical requirements to bear in mind as they cultivate their data protection strategy for their Kubernetes environments.

10 Mins Break | Turnaround | Raffle x 2

Speaker: Michael “MJ” Johnson – MinIO

Abstract: The interplay between Kubernetes and storage has never been particularly clean. Container Storage Interface (CSI) drivers improved upon plugins but still did not provide a clear path to the drives. The Container Object Storage Interface brought cloud-native interfaces to the equation but did not solve for the custom driver challenge. LocalPV provided direct access to the drives but lacked the sophistication of the CSI driver.

The answer lies with DirectPV. DirectPV is an open source implementation that provides direct access to the persistent volumes ensuring that databases, data stores and data processing workloads have the simplest and most performant path to the underlying data. It is defined as much by what it does not do (erasure coding, replication) as by what it does (drive discovery, provisioning, affinity, isolation, monitoring and management).

At the most basic level, DirectPV is a distributed persistent volume manager, and not a storage system. As a result, DirectPV does not introduce extra layers of replication/erasure coding and extra network hops in the data path.

In this talk, Michael “MJ” Johnson, will introduce DirectPV, its components and will outline where it should be used.

5 Mins Break | Turnaround | Raffle x 2

Speaker: Shivay Lamba & Rishit Dagli 

Abstract: SODA Foundation’s Multicloud Platform enables businesses to set up hybrid/inter and intra cloud platforms for their use cases.

Traditionally companies trying to set up a cloud infrastructure might want to experiment and choose one cloud provider for all their use cases. But in most cases it is not the most optimized and cost effective approach.

This talk aims to enlighten these developers and business owners to make them understand about hybrid/inter cloud strategies and how it can be more cost effective using Artificial Intelligence ( cost forecasting based on API invocation and computation/storage) and more scalable instead of using a single cloud platform or struggling to find one. The talk will also cover how the SODA Multicloud Project is used to provide this multi cloud ability.

So the talk will have the following pointers :

1. It will begin with an introduction to different types of cloud architectures ( hybrid, inter, intra)
2. Talk about how SODA Multicloud Architecture Works
3. Integrating SODA Multicloud with general cloud use cases.