Storage Spaces and Windows as a Storage Solution
This section briefly covers the big shift in storage introduced with Windows Server
- Although it’s not strictly a Hyper-V topic, I cover it because it will most likely
affect the way your Hyper-V environments are architected, especially in smaller
organizations and branch offices.
In the introduction, I talked about Fibre Channel and iSCSI-connected SANs, which
historically have been the preferred storage choice for organizations because they
provide many benefits:
Storage is centralized, allowing the highest utilization of the available space as
opposed to many separate instances of storage with lots of wasted space.
Centralized backup is possible.
They offer the highest level of performance and scalability, which is possible
because the storage is all centralized, allowing higher-specification storage
solutions to be purchased.
Storage is accessible to all servers throughout the datacenter, provided the server
has the required connectivity, such as Fibre Channel or iSCSI.
Centralized storage enables easy migration of virtual machines between physical
hosts, because the storage can be seen by multiple servers.
They provide shared storage, which is required for many cluster scenarios.The use of high-end, centralized storage is still a great option for many organizations
and scenarios. However, another model is also being adopted by many organizations
and service providers, including Microsoft Azure: just a bunch of disks (JBOD)
solutions that are either local to a server or in an external enclosure connected to
numerous clustered hosts. This can provide great cost efficiencies because storage
subsystems based on regular disks are much cheaper than SAN solutions. It is
important, though, to build in resiliency and backup solutions for what is now local
storage containing critical workloads (that is, your virtual machines).
Windows Server has long had the ability to create fault-resilient storage using
redundant array of independent disks (RAID) technology in one of two modes: RAID-
1, which mirrored all data from one disk to another disk, and RAID-5, which used
striping with parity. However, there were challenges with the software RAID
implementation:
It did not self-heal. If a disk was lost, another disk had to be configured manually
to be the replacement.
Only thick/fat provisioning was possible, which means a volume can be created
only up to the physical space available. Thin provisioning, which allows volumes to
be created beyond the physical space and allocated as data was written, was not