still the focus filesystem for Storage Spaces in Windows Server 2016, outside of
Storage Spaces Direct.
STORAGE SPACES DIRECT
I’ve mentioned Storage Spaces Direct (S2D) a lot so far because it’s a game changer for
storage with Windows Server 2016. Storage Spaces Direct builds on the idea of more
efficiently using direct-attached storage and extends this to cluster scenarios with a
pure software-defined storage solution. Storage Spaces Direct uses the local storage in
servers, which could be internal disks or in an external storage enclosure connected
via SAS HBA or SATA, and aggregates the storage to make it available as a Cluster
Shared Volume providing high availability for the data and providing a scalable
solution. Storage Spaces Direct uses industry-standard hardware and the existing
Ethernet or RDMA network as the storage fabric for the internal replication and
coordination of aggregating the local disks to a cluster volume over SMB 3. RDMA is
recommended for the network connectivity because of its low server overhead and
higher bandwidth connections compared to regular Ethernet, which is enabled
automatically when present via SMB Direct. Figure 4.6 shows a high-level view of
utilizing storage with Storage Spaces Direct.
Figure 4.6 High-level view of Storage Spaces Direct utilizing SMB 3 to aggregate local
storage for nodes in a cluster
Storage Spaces Direct is used in two models: hyperconverged and converged. In a
hyperconverged model, both the compute and storage resources are the same nodes
in the same cluster; that is, the same nodes that have the local storage used by Storage
Spaces Direct are also hosting the VMs that use that storage. In this model, the
compute and storage scale together and are managed together. If more storage is
needed, you have to add nodes, which would also increase your compute scale even if
this is not required, and vice versa. Because of this tight coupling of scale, the
hyperconverged deployment model tends to be used in small-to-medium sized
deployments. In the converged model, one cluster is used for the storage services and
offers the CSVs via Scale-out File Servers (SoFS), and a separate cluster is used for
Hyper-V and the storage is accessed using SMB 3. This deployment model is used for
larger environments, because the compute and storage can scale separately in addition
to being managed separately.