Cephalocon 2024

Deutsche Telekom operates a growing email system with several million accounts and billions of emails stored on traditional NFS. Six years ago we introduced librmb (librados mailbox) to the community, a universal open source library to store emails in a Ceph cluster. Librmb uses RADOS to store email directly in Ceph to achieve maximum performance through parallel access from many email gateways simultaneously, for millions of active customers. Email systems are much too complex to be simulated in a way which would allow to verify if the switch to librmb will work for a large number of users. Therefore a field test with active customers was necessary to provide an educated guess on the behavior of the final setup. This presentation will cover the results from artificial and real field tests with more than 1 million accounts/users. The results include the experience and learnings of migrating from the existing email system into Ceph, an extended time of running the system and from migrating the accounts out of the test system. We will provide an insight into our learnings, found issues, potential solutions and an outlook into our next steps towards a Ceph based email system.

The Crimson project is an effort to build a replacement ceph-osd well suited to the new reality of low latency, high throughput, persistent memory and NVMe technologies. Built on the Seastar C++ framework, crimson-osd aims to be able to fully exploit modern devices by minimizing latency, CPU overhead, and cross-core communication. This talk will discuss the current state of Crimson going into the Tentacle release.

Pawsey receives infrequent blocks of government funding to provide supercomputing, cloud and data services typically at no cost to scientific researchers in Australia. This has a big impact on both the way requirements are determined and on the design and procurement model. Our new 20PB cluster pushes the bounds of what is recommended, for performance and for node density, but we are confident it will serve researchers' needs over the lifespan of the equipment and beyond. This talk will look at the details of this design, and how we determined them, tested them, and increased our confidence of success. Many of these factors are useful in deploying any new cluster, so you will take away some tips on how to get started, or use the experience you already have to build the next one.

End to end tracing can help debug latency issues between the RGW and the OSD, giving a complete picture of request flow, but tracing itself has a performance impact on the system. When your production system runs into issues, the last thing you want is to put it under more pressure! In this session we will demonstrate how to use Lua scripting on the RGW to turn opentelemetry based tracing only for some of the incoming requests. Allowing us to focus on the problem without slowing down the entire system.

The Ceph project has used the Teuthology testing framework for much of its history. The custom framework is used to schedule batch jobs that perform e2e testing of Ceph. This is orchestrated using a suite of YAML fragments to alternate test modes, configurations, workloads, and other parameters. Teuthology assembles these fragments into a static matrix with potentially dozens of dimensions ultimately producing a combinatoric explosion of jobs which are evaluated, in practice, as smaller subsets for scheduling. We will explore an alternative directed graph model for constructing jobs from a suite of YAML fragments using path walks. Code adapted to this model has been constructed to produce subsets in linear time and provide Lua scriptable control of YAML fragment generation. The latter new feature empowers us to test Ceph with more rigor and completeness. For example, upgrade suites can be constructed using all possible versions of Ceph that are valid upgrade paths to a target release. We will explore this and other enhancements in depth. The audience can expect to leave with a firm and visual understanding of how QA is performed on Ceph and a vision for the future testing.

Ceph supports a wide range of Erasure Coding plugins and techniques. This talk will give a brief overview of the different erasure coding plugins supported by Ceph along with some interesting detail about how the various techniques for each plugin works. After giving an overview of the different plugins and techniques, we will discuss our motivations for changing the default plugin from Jerasure to ISA-L, including highlighting some of the more recent optimisations supported by ISA-L. We will also show some preliminary performance results supporting this decision.

LINE, a global mobile messenger, has adopted Ceph as its main object storage. It is used to store different kinds of data, such as log files and application data. Thanks to its scalability, billions of objects are stored in our clusters. However, over time, object deletions lead to the accumulation of tombstones in RocksDB, resulting in delays during iteration. Slow iteration not only impacts LIST operation but also stalls subsequent requests. To address this issue, we first collected RocksDB metric called "skip count", which indicates the total number of tombstones detected during iterations. We then deployed a new job which compacts OSDs with high skip counts to prevent stalls. Additionally, we analyzed the pattern of tombstones and found out that a few prefixes account for over 80% tombstones, throughout the entire OSD. Based on this observation, we propose range-based compaction. In this presentation, we will first explain the basics of RocksDB and its role in Ceph Object Storage. Then, we will share our experience how we handled the RocksDB issue. Lastly, we will discuss our proposal for range-based compaction, which could further optimize overall system performance.

As new generations of hardware become available and Ceph is improved, how does it's performance change? If we look back 4 years, how did Ceph's performance improve (or not)?

Ceph, the words most popular open source software defined storage system, has offered storage efficiency features such as block device compression, object compression and server-side object encryption for a number of releases. However, enabling these features has always come as a trade-off between the additional performance required (in terms of cores/GHz) vs the raw storage cost, ultimately driving users away from these features. In this talk we will walk through several different scenarios where Intel's QAT offload is used to enable these features without significant overhead to primary processing, and still yields greater performance without causing increased cost per GB.