Book Image

Ceph: Designing and Implementing Scalable Storage Systems

By : Michael Hackett, Vikhyat Umrao, Karan Singh, Nick Fisk, Anthony D'Atri, Vaibhav Bhembre
Book Image

Ceph: Designing and Implementing Scalable Storage Systems

By: Michael Hackett, Vikhyat Umrao, Karan Singh, Nick Fisk, Anthony D'Atri, Vaibhav Bhembre

Overview of this book

This Learning Path takes you through the basics of Ceph all the way to gaining in-depth understanding of its advanced features. You’ll gather skills to plan, deploy, and manage your Ceph cluster. After an introduction to the Ceph architecture and its core projects, you’ll be able to set up a Ceph cluster and learn how to monitor its health, improve its performance, and troubleshoot any issues. By following the step-by-step approach of this Learning Path, you’ll learn how Ceph integrates with OpenStack, Glance, Manila, Swift, and Cinder. With knowledge of federated architecture and CephFS, you’ll use Calamari and VSM to monitor the Ceph environment. In the upcoming chapters, you’ll study the key areas of Ceph, including BlueStore, erasure coding, and cache tiering. More specifically, you’ll discover what they can do for your storage system. In the concluding chapters, you will develop applications that use Librados and distributed computations with shared object classes, and see how Ceph and its supporting infrastructure can be optimized. By the end of this Learning Path, you'll have the practical knowledge of operating Ceph in a production environment. This Learning Path includes content from the following Packt products: • Ceph Cookbook by Michael Hackett, Vikhyat Umrao and Karan Singh • Mastering Ceph by Nick Fisk • Learning Ceph, Second Edition by Anthony D'Atri, Vaibhav Bhembre and Karan Singh
Table of Contents (27 chapters)
Title Page
About Packt
Contributors
Preface
Index

Promotion throttling


As has been covered earlier, promotions are very expensive operations in tiering and care should be taken to make sure that they only happen when necessary. A large part of this is done by carefully tuning the HitSet and recency settings. However, in order to limit the impact of promotions, there is an additional throttle that restricts the number of promotions to a certain speed. This limit can either be specified as number of bytes or objects per second via two OSD configuration options:

    osd_tier_promote_max_bytes_sec

    osd_tier_promote_max_objects_sec

The default limits are 4 MBps or five objects a second. While these figures may sound low, especially when compared with the performance of the latest SSDs, their primary goal is to minimize the impact of promotions on latency. Careful tuning should be done to find a good balance on your cluster. It should be noted that this value is configured per OSD, and so the total promotion speed will be a sum across all OSDs...