Ceph Days Asia 2024

An update from the leaders of the Ceph project about recent developments, current priorities, and other activity in the Ceph community, including what's new in the latest Squid release, and what is coming up for the next major release, Tentacle.

In this talk, we will explore Rook, its strengths, and how it compares to Cephadm. We'll address the question, "Is Rook a good fit for you?" by examining its deployment in large-scale production environments and its integration with existing clusters.

We will discuss benchmarking and optimizing your Rook cluster for production, along with monitoring and troubleshooting strategies. Additionally, we will look at potential future improvements in Rook Ceph.

This session offers a concise overview of our journey from zero to a fully operational Rook Ceph cluster, sharing key learnings and insights along the way.

In this talk, we will explore NVMe-oF Recovery, which enables fast recovery through NVMe-oF-based storage disaggregation. In large-scale distributed storage systems, compute node failures are common. These failures may trigger unexpected I/Os to restore system availability. NVMe-oF offers an opportunity to disaggregate compute and storage resources, reducing the impact of such failures. However, Ceph's Converged Architecture tightly couples compute and storage resources, so a compute failure brings storage down as well. This forces the storage to undergo recovery, even if it's still usable. To overcome this, we will introduce NVMe-oF Recovery, a solution that ensures fast recovery by separating compute and storage failure domains within the CRUSH Map, while maintaining the reliability and scalability of traditional systems.

Data processing large amount of data on traditional storage comsumes large amount resources. Offloading computation to object storage provides several benefits to users. This session briefly intoduce typical data processing, what is Apache Arrow and how it can help users with Ceph.

Sage is a security analyst at Red Hat, focusing on storage security, both for containers and for distributed systems. They did a master's degree at UCSC, and have an undergraduate degree in math and computer science from Umass Amherst. In their free time, they enjoy hiking with their dog.

This talk will explore why Ceph is the best software-defined storage solution available. By examining its technical evolution from 2004 until now, we will highlight the key points that demonstrate its leadership over today's alternatives.

Deploying and operating a Ceph cluster is complex because Ceph is designed to be a general-purpose storage solution. MicroCeph is an open source Ceph orchestration tool which makes it easy to deploy a Ceph on a single node and scale it to a highly available cluster as needed. This reduces the barrier of entry and makes Ceph more accessible to everyone but specially to the new users for training. For operations, MicroCeph makes configuration, scaling CRUSH, spawning services across cluster even easier. The intention of this talk would be to introduce MicroCeph to the viewers and promote the practices it employs for Ceph.

Ceph is a distributed software-defined storage with scalability, delivering block, file and object in one unified system. It has been used in Cloud Stacks such as OpenStack and Kubernetes. We have been using Ceph clusters for VM image store, Kubernetes Persistent Volumes, NFS servers, iSCSI.

In this talk, we will present how to deploy a new version of a Ceph cluster in four different environments

• New deployment with Internet Connection
• Upgrade from an old version
• New deployment in an air-gapped environment
• Upgrade from an old version of a Ceph in an air-gapped environment

The object store provided by RGW has a weakness in that the object listing time increases depending on the number of objects. We share the results and analysis of the performance experiment conducted to prepare a countermeasure to this weakness. And we also share the derived tips for use.

- Ceph의 데이터 정합성을 체크하기 위한 scrub, deep-scrub 집중적으로 분석
- 대부분 scrub / deep-scrub 은 default 로 사용 가능하지만, 대용량과 파일 수가 많은 경우 수행 지연으로 인해 정상적인 운영이 어려운 경우 분석
- scrub / deep-scrub 처리 과정을 이해하고 상황과 환경에 따라 어떻게 하면 보다 효율적으로 운영할 수 있는지에 대한 연구 결과 및 실적용 아이디어를 공유