Cloud Native Brand New Day

1. 6:00 PM - 6:40 PM +0530

   A Door is Not a House and a Portal is not a Platform

   in-person

   Internal Developer Platforms (IDPs) and Internal Developer Portals (IDPs) are often used interchangeably, but they solve different problems. Understanding the distinction is important to building a better developer experience on Kubernetes.

   In this talk, we’ll explore the relationship between developer platforms, developer portals, and the capabilities that enable teams to build, deploy, and operate applications. We’ll look at how tools such as Backstage fit into the picture, why a developer portal alone does not make a platform, and how platform capabilities can be exposed through different interfaces, including portals, CLIs, and APIs.

   Using practical examples and concepts from the cloud-native ecosystem, we’ll discuss golden paths, self-service workflows, and the role of platform abstractions in reducing unnecessary complexity for application developers.

   Attendees will leave with a clearer mental model of portals versus platforms, a better understanding of where each fits into a Kubernetes-based environment, and practical guidance for identifying developer pain points before choosing tools or building platform capabilities.

   SPEAKERS
2. 6:40 PM - 7:20 PM +0530

   Proactive Autoscaling for LLM Workloads - A context-aware predictive approach on Kubernetes

   in-person

   Large Language Model (LLM) workloads introduce unique challenges for autoscaling on Kubernetes. Unlike traditional applications, LLM inference workloads are influenced by factors such as prompt length, output token generation, request complexity, and varying response times. Reactive autoscaling approaches that rely on metrics such as CPU and memory utilization may struggle to respond quickly enough to sudden changes in demand, leading to increased latency, inefficient resource utilization, and higher infrastructure costs.

   In this talk, we’ll explore a context-aware, predictive approach to autoscaling LLM workloads on Kubernetes. We’ll discuss how workload characteristics, request patterns, and inference-related metrics can be used to anticipate resource demand and make scaling decisions before bottlenecks occur. We’ll also examine the limitations of traditional autoscaling mechanisms and how predictive strategies can help balance responsiveness, performance, and resource efficiency.

   Through practical concepts and cloud-native approaches, attendees will gain insights into designing smarter autoscaling strategies for LLM inference workloads, understanding the signals that matter beyond CPU and memory, and applying predictive scaling techniques to improve the reliability and efficiency of AI applications running on Kubernetes.

   SPEAKERS
3. 7:20 PM - 8:00 PM +0530

   Case Study: Unlimited Log Retention and Searchability Using EFS, AWS DataSync, S3 Glacier IR, and Athena in Grafana

   in-person

   Keeping application logs for months or years is easy to promise and expensive to deliver. Our observability platform holds 30 days of logs, so anything older meant finding a file in S3, downloading gigabytes by hand and searching it in a code editor, which was slow when an incident was already under way.

   This case study walks through how we fixed that on AWS and EKS. Every production service writes logs through its Log4j configuration to a shared EFS volume mounted into each namespace. An AWS DataSync job copies those files into S3, where they move to Glacier Instant Retrieval. That keeps about 2.4 TiB for roughly a tenth of the cost of S3 Standard. Athena then queries the files in place, with no re-ingest and no restore wait, and the Grafana we already run on EKS turns the results into a searchable log view with a volume chart and a time-range picker.

   We’ll cover the architecture, the alternatives we compared and why we dropped them, and the real cost numbers: about five cents to search a day of our busiest service, and around $5 to $20 a month for typical team use. We’ll also share the pitfalls, including file-naming patterns, scan limits that stop a bad query from becoming an expensive one, and keeping access keyless with IAM roles.

   You’ll learn

   How to build a low-cost, long-term log pipeline on EFS, DataSync and S3 How to query cold logs in place with Athena, and what actually drives the cost How to expose the results in an existing Grafana without adding a new platform Guardrails that keep a pay-per-query setup safe for a whole team

   SPEAKERS

No sessions match the selected labels.