Event Details

We deployed a nearly 200 nodes OpenStack environment, in order to more easily manage virtual disks, all the instances are booting from volumes. We use Rally to evaluate the performance of this cluster. The fact is, compared to boot from local disks, booting from volumes often lead to failure.These failure are largely due to the low performance of Cinder. In this presentation, we will discuss problems found during performance testing in detail.Database, threading model, deployment architecture and other factors will affect the performance of cinder largely. Under certain conditions all of these can become a performance bottleneck. We will introduce these separately.

Finally, we will give our conclusions and solutions in order to help everyone to be more convenient to use cinder.