13.2. Primary Storage

This section gives concepts and technical details about CloudStack primary storage. For information about how to install and configure primary storage through the CloudStack UI, see the Installation Guide.

Section 2.6, “About Primary Storage”

13.2.1. Best Practices for Primary Storage

The speed of primary storage will impact guest performance. If possible, choose smaller, higher RPM drives or SSDs for primary storage.
There are two ways CloudStack can leverage primary storage:
Static: This is CloudStack's traditional way of handling storage. In this model, a preallocated amount of storage (ex. a volume from a SAN) is given to CloudStack. CloudStack then permits many of its volumes to be created on this storage (can be root and/or data disks). If using this technique, ensure that nothing is stored on the storage. Adding the storage to CloudStack will destroy any existing data.
Dynamic: This is a newer way for CloudStack to manage storage. In this model, a storage system (rather than a preallocated amount of storage) is given to CloudStack. CloudStack, working in concert with a storage plug-in, dynamically creates volumes on the storage system and each volume on the storage system maps to a single CloudStack volume. This is highly useful for features such as storage Quality of Service. Currently this feature is supported for data disks (Disk Offerings).

13.2.2. Runtime Behavior of Primary Storage

Root volumes are created automatically when a virtual machine is created. Root volumes are deleted when the VM is destroyed. Data volumes can be created and dynamically attached to VMs. Data volumes are not deleted when VMs are destroyed.

Administrators should monitor the capacity of primary storage devices and add additional primary storage as needed. See the Advanced Installation Guide.

Administrators add primary storage to the system by creating a CloudStack storage pool. Each storage pool is associated with a cluster or a zone.

With regards to data disks, when a user executes a Disk Offering to create a data disk, the information is initially written to the CloudStack database only. Upon the first request that the data disk be attached to a VM, CloudStack determines what storage to place the volume on and space is taken from that storage (either from preallocated storage or from a storage system (ex. a SAN), depending on how the primary storage was added to CloudStack).

13.2.3. Hypervisor Support for Primary Storage

The following table shows storage options and parameters for different hypervisors.

	VMware vSphere	Citrix XenServer	KVM
Format for Disks, Templates, and Snapshots	VMDK	VHD	QCOW2
iSCSI support	VMFS	Clustered LVM	Yes, via Shared Mountpoint
Fiber Channel support	VMFS	Yes, via Existing SR	Yes, via Shared Mountpoint
NFS support	Y	Y	Y
Local storage support	Y	Y	Y
Storage over-provisioning	NFS and iSCSI	NFS	NFS

XenServer uses a clustered LVM system to store VM images on iSCSI and Fiber Channel volumes and does not support over-provisioning in the hypervisor. The storage server itself, however, can support thin-provisioning. As a result the CloudStack can still support storage over-provisioning by running on thin-provisioned storage volumes.

KVM supports "Shared Mountpoint" storage. A shared mountpoint is a file system path local to each server in a given cluster. The path must be the same across all Hosts in the cluster, for example /mnt/primary1. This shared mountpoint is assumed to be a clustered filesystem such as OCFS2. In this case the CloudStack does not attempt to mount or unmount the storage as is done with NFS. The CloudStack requires that the administrator insure that the storage is available

With NFS storage, CloudStack manages the overprovisioning. In this case the global configuration parameter storage.overprovisioning.factor controls the degree of overprovisioning. This is independent of hypervisor type.

Local storage is an option for primary storage for vSphere, XenServer, and KVM. When the local disk option is enabled, a local disk storage pool is automatically created on each host. To use local storage for the System Virtual Machines (such as the Virtual Router), set system.vm.use.local.storage to true in global configuration.

CloudStack supports multiple primary storage pools in a Cluster. For example, you could provision 2 NFS servers in primary storage. Or you could provision 1 iSCSI LUN initially and then add a second iSCSI LUN when the first approaches capacity.

13.2.4. Storage Tags

Storage may be "tagged". A tag is a text string attribute associated with primary storage, a Disk Offering, or a Service Offering. Tags allow administrators to provide additional information about the storage. For example, that is a "SSD" or it is "slow". Tags are not interpreted by CloudStack. They are matched against tags placed on service and disk offerings. CloudStack requires all tags on service and disk offerings to exist on the primary storage before it allocates root or data disks on the primary storage. Service and disk offering tags are used to identify the requirements of the storage that those offerings have. For example, the high end service offering may require "fast" for its root disk volume.

The interaction between tags, allocation, and volume copying across clusters and pods can be complex. To simplify the situation, use the same set of tags on the primary storage for all clusters in a pod. Even if different devices are used to present those tags, the set of exposed tags can be the same.

13.2.5. Maintenance Mode for Primary Storage

Primary storage may be placed into maintenance mode. This is useful, for example, to replace faulty RAM in a storage device. Maintenance mode for a storage device will first stop any new guests from being provisioned on the storage device. Then it will stop all guests that have any volume on that storage device. When all such guests are stopped the storage device is in maintenance mode and may be shut down. When the storage device is online again you may cancel maintenance mode for the device. The CloudStack will bring the device back online and attempt to start all guests that were running at the time of the entry into maintenance mode.