Storage Virtualization

Storage virtualization maps logical storage to physical storage through an abstraction layer.

S3 covers SAN, NAS, storage arrays, and LUNs. Virtualization pools these resources and presents logical volumes independently of individual storage devices.

Storage Pools

• Multiple supported storage arrays contribute capacity to a pool.
• Logical volumes are created from the pool and assigned to compute systems.
• Mapping hides the physical location of stored blocks or files.
• Central management simplifies allocation across heterogeneous devices.
• Supported implementations can migrate data or change capacity without interrupting client access.

Pooling can increase utilization and reduce total cost of ownership. Non-disruptive operation depends on the implementation and available capacity.

Storage for VMs

A VM commonly consists of configuration files, virtual disk files, and optional saved-state files.

• A virtual disk presents a block device to the guest OS.
• The hypervisor maps guest disk operations to the backing storage.
• The guest sees its assigned disk capacity rather than the entire storage pool.
• A virtual disk can be backed by local storage, SAN storage, or a NAS datastore.
• Provisioned virtual capacity can differ from physical space consumed.

Shared Datastores

VMFS

Virtual Machine File System (VMFS) is VMware’s clustered filesystem for storing VM files on block storage.

• Multiple ESXi hosts can access a shared VMFS datastore.
• Locking coordinates access to VM files.
• A datastore can hold virtual disks belonging to many VMs.

Shared datastore access does not permit arbitrary concurrent writes to the same virtual disk. VMware’s locking behaviour protects disks from conflicting access.

NFS

Network File System (NFS) provides remote file access over a network. A hypervisor can store VM files in an NFS export supplied by a NAS system.

• The NAS system manages the underlying filesystem.
• Multiple authorized hosts can access the export.
• Network availability and storage performance affect VM I/O.

VMFS and NFS are the lecture’s datastore examples. VMFS is not a generic filesystem used by every hypervisor.

Block and File Virtualization

A network-based virtualization appliance can perform the mapping between clients and multiple storage systems. A global file namespace allows files to move between NAS systems while retaining client-visible paths.

Thick and Thin Provisioning

• Thick provisioning
  Reserves backing capacity for the provisioned volume in advance.
• Thin provisioning
  Allocates backing capacity as data is written.

Thin provisioning leaves unwritten capacity available in the shared pool. It does not create physical capacity. Pool exhaustion can cause allocation or write failures, so growth requires monitoring.

Worked Example

The lecture uses a 2 TB storage pool and 3 LUNs. Use decimal units, with 2 TB equal to 2000 GB. Ignore metadata and redundancy overhead.

1. Total provisioned capacity is $500 + 550 + 800 = 1850$ GB.
2. Actual data occupies $100 + 50 + 200 = 350$ GB.
3. Thick provisioning leaves $2000 - 1850 = 150$ GB unallocated in the pool.
4. Thick provisioning reserves $1850 - 350 = 1500$ GB that contains no data.
5. Thin provisioning initially consumes 350 GB, leaving $2000 - 350 = 1650$ GB available in the pool.

Both arrangements present the same 1850 GB of logical capacity. Thin provisioning makes the 1500 GB of unwritten capacity available for allocation elsewhere until writes consume it.