many small data devices or a few large ones

Latest response

I'm working on re-laying out my RHEV environment to clean up years of mis-management. Would it be better to recreate the storage domains with many small volumes, 1TB or so, or a few large volumes, 5-10TB?


What are the characteristics of your environment? How many hosts, templates and Vm's? What is the average size and lifetime of templates and VM's?

In our Continuous Integration setup, we have many, short-living Vm's and many, large templates. We experienced bottlenecks in the RHEVM-SPM communication and scaled by creating multiple RHEVM instances each with 8 hypervisors and a single virtual data center, cluster and storage domain.

With long-living VM's and identical performance requirements, the RHEVM-SPM will not be a bottleneck and a single data center, cluster and storage domain will be easier to manage. When you have different performance requirements, you could create for example an SSD based storage domain for your high-end VM's.

Our environment grew out of control for a while, now I have been asked to jump in and clean up so I am still familiarizing myself with it all. While this is an existing environment that I am cleaning up, I am looking at it more as a clean slate.

The VMs tend to have a 35GB disk footprint for the OS, however some are up to 100GB. Generally the VMs have 8GB or less RAM and 4 cores but a few are up to 48GB with 16 cores.

In all clusters combined we have around 500. The largest cluster has over 250 VMs with 26 hypervisors, this is my problem child. That Data Center has 2 other smaller clusters, all 3 clusters live on 7 storage domains ranging from 300 GB to 5TB. There is a second DC with several smaller clusters.

The templates generally are only useful for 6 months but we have older ones available. In the long term there will be more templates allowing me to be more specific with the function and therefore the templates will be much smaller.

The VMs tend to stick around for quite some time although my task of cleaning up will result in more, smaller VMs with a much shorter lifespan.