Figure 4.7 shows the two deployment choices. Typically, you should have a minimum
of four nodes to use parity resiliency and/or multiresiliency options. However, a
three-node cluster and two-node cluster are possible with mirror resiliency providing
resiliency from a single-node failure and one or two disk failures, depending on
whether two-copy or three-copy mirroring is used, respectively.
Figure 4.7 The two types of deployment model for Storage Spaces Direct
The network provides the bus over which the data is replicated to the multiple targets.
The data is sent to the nodes synchronously and written to durable storage before the
write is acknowledged to the application, so a high-speed and reliable network
between nodes is critical. 10Gbps should be considered the minimum speed available
between nodes, with preferably RDMA network connectivity. There is no hard block
from a network less than 10Gbps,but performance would suffer and not be suitable
for production. When thinking of latency between the nodes, you should be thinking
in terms of nanoseconds, not milliseconds.
Behind the scenes, Storage Spaces Direct works via the Software Storage Bus (SSB),
which brings together all of the local storage across the cluster and makes it behave as