Book Image

Learning Ceph - Second Edition

By : Karan Singh, Vaibhav Bhembre, Anthony D'Atri
Book Image

Learning Ceph - Second Edition

By: Karan Singh, Vaibhav Bhembre, Anthony D'Atri

Overview of this book

Learning Ceph, Second Edition will give you all the skills you need to plan, deploy, and effectively manage your Ceph cluster. You will begin with the first module, where you will be introduced to Ceph use cases, its architecture, and core projects. In the next module, you will learn to set up a test cluster, using Ceph clusters and hardware selection. After you have learned to use Ceph clusters, the next module will teach you how to monitor cluster health, improve performance, and troubleshoot any issues that arise. In the last module, you will learn to integrate Ceph with other tools such as OpenStack, Glance, Manila, Swift, and Cinder. By the end of the book you will have learned to use Ceph effectively for your data storage requirements.
Table of Contents (18 chapters)
Title Page
Credits
About the Authors
About the Reviewer
www.PacktPub.com
Customer Feedback
Preface

Client settings


Ceph's RBD block service is heavily utilized by cloud platforms like OpenStack and other virtualization solutions. Effective caching in memory on the client side can significantly improve the performance that users experience:

[client.yourclientname]
rbd_cache = true
rbd_cache_size = 67108864
rbd_cache_max_dirty = 50331648
rbd_cache_target_dirty = 33554432

These values of 64MB, 48MB, and 32MB are double the default numerical values as of Luminous.

Note

Settings must be selected so thatrbd_cache_size > rbd_cache_max_dirty > rbd_cache_target_dirty.

Some references suggest larger values, but consider that these cache buffers are allocated per attached volume. Consider a dense hypervisor node hosting 100 smaller VM instances, each mapping an RBD volume for its boot drive. With these values, the hypervisor node will dedicate more than 6 GB of RAM just for caching guest boot drives. You may be able to spare that much, but hypervisors are often limited by RAM, and this should be...