Presentations
In cloud-native environments, application performance often degrades due to contention over shared resources such as CPU caches and memory bandwidth. Current container technologies do not isolate these resources, which compels operators to maintain low utilization by scaling out their deployments.
This session will show how memory noisy neighbor degrades application performance, survey strategies used in production by Google and Alibaba Cloud to mitigate such performance interference, and discuss community efforts, highlighting the memory-collector observability project.
Over the past 3 years I have evaluated the scaling capabilities of various applications. I will share my learnings about application performance, enabling you to understand and scale your applications effectively. Attendees will learn how to:
- Proactively prepare for gradual and abrupt system stress
- Assess and confirm their application’s scaling potential
- Identify performance bottlenecks
- Implement modifications efficiently
- Eliminate blind spots in stress testing and scalability validation
As vector databases and Generative AI gain momentum, this session introduces the *MyVector Plugin*, which enables vector storage and similarity search in MySQL. Using MySQL’s server component plugin and UDF, the *MyVector Plugin* integrates vector search into MySQL + InnoDB without requiring a separate vector database. The talk covers technical details, challenges, and practical applications, offering valuable insights on adding vector search to MySQL systems and covering how vectors work, practical applications, or both.
In this talk, Murriel and Elizabeth will be your guides on a brief tour of several open source tools for deploying a workload into Kubernetes. Our journey will begin with manual hello world deployments and from there we will explore some of the most common modern tools for CI/CD, including a demo speedrun!
Introduction to fully open source alternative for MongoDB. Show how easily you can replace expensive and vendor-locked-in MongoDB services using PostgreSQL in the Kubernetes cluster + FerretDB.
Per-CPU variables are widely used in the Linux kernel to reduce contention of atomics and locks in large scale systems. This makes them an attractive API to use from Rust, but exposing this API requires interacting with all levels of the kernel stack: from the build system and linker all the way to arch-specific optimizations and runtime allocation of new per-CPU variables. This talk is intended for an audience that is broadly familiar with kernel primatives and synchronization issues but entirely new to Rust.
A behind-the-scenes look at how our small systems engineering team manages our infrastructure. This is centered around MySQL and MongoDB.
We'll go through real-world implementation with code examples, showing how we:
- Built a flexible automation framework that adapts to new technologies
- Created a sustainable way for a small team to manage enterprise-scale infrastructure
- Turned "those #@$% servers" into well-behaved, predictable systems
Ceph is an open source distributed object store, network block device, and file system designed for reliability, performance, and scalability. It runs on commodity hardware, has no single point of failure, and is supported in the Linux kernel.
This tutorial will describe the Ceph architecture, share its design principles, and discuss how it can be part of a cost-effective, reliable cloud stack.
Explore the evolving integration of CephFS with OpenStack Manila in this session, where we’ll delve into the unique features CephFS brings to cloud file systems. We’ll cover the trade-offs of CephFS Native and CephFS NFS, highlighting practical reasons why each might not fit all use cases. Wrapping up, we’ll look at future innovations, like the SMB driver and VirtIOFS, that promise to expand functionality and use cases for OpenStack file sharing. Join us for a deep dive into these powerful tools and their future in cloud architecture.
This talk will focus on the development of an open-source Earth System Model (ESM) by the Climate Modeling Alliance (CliMA), which aims to address the limitations of existing, outdated ESMs that are computationally expensive and not readily publicly accessible. CliMA's model is being built from scratch to be less resource-intensive while maintaining scientific accuracy, making it more useful to the broader scientific community. The talk will highlight the benefits of providing such a model as open source to support climate research and collaboration.