4. Resource Group Architecture : 4.3 Storage Resources
4.3 Storage Resources
Designing storage resources for vCloud differs from the traditional vSphere approach. Platform features such as VMware vSphere Storage DRS™ and storage profiles assist in balancing workloads across storage resources, allowing providers to offer differentiated storage. This allows the provisioning of flexible pools of storage resources while maintaining consistent performance for end users. Users can choose the right storage tier for a particular type of workload.
VMware recommends the following when designing storage resources:
*Perform a current state analysis for storage usage and trends.
*Define the range of storage SLAs needed and appropriate pricing models.
*Create multiple storage profiles in vSphere, based on SLAs, workloads, and cost.
*Map storage profiles to the provider virtual datacenter.
*Select a subset of storage profiles provided by the provider virtual datacenter to the organization virtual datacenter.
*Design for optimal availability (redundant paths from vSphere hosts to storage fabric).
*Deploy modular and scalable physical storage.
*Monitor storage usage and trends using capacity analysis tools.
*Use storage performance tools to tune vApp storage workloads.
vCloud Director supports tiering storage within a virtual datacenter using storage profiles. Configure shared storage and storage profiles in the resource groups per vSphere design guidelines.
Datastore sizing considerations include both capacity and performance:
*Datastore capacity considerations:
*What is the optimal size for datastores based on the physical storage and vCloud workload expectations?
*What is the average vApp size x number of vApps x spare capacity? For example,
average virtual machine size x number of virtual machines x (1+ % headroom).
*What is the average virtual machine disk size?
*On average, how many virtual machines are in a vApp?
*What is the expected number of virtual machines?
*How much reserved spare capacity is needed for growth?
*Will expected workloads be transient or static?
*Is fast provisioning used?
*Datastore performance considerations:
*Will expected workloads be disk intensive?
*What are the performance characteristics of the associated cluster?
NotevCloud Director does not support Raw Device Mapping (RDM).