Netherlands: Software

Introductie Windows Server 2016

Issue link:

Contents of this Issue


Page 75 of 173

67 C H A P T E R 3 | Storage Note You can use the Test-SRTopology Windows PowerShell cmdlet to ensure that you meet the requirements and assist with recommended configuration tuning for log files. Scenarios Storage Replica was designed with two scenarios in mind: Stretching of a failover cluster for high availability Replication between servers for disaster recovery Stretch cluster replication A stretch cluster (also referred to as a multisite cluster) uses Storage Replica to connect two sets of asymmetric shared storage within a single failover cluster. This storage can be serially attached SCSI JBOD (just a bunch of drives), iSCSI target, or SAN. Cluster nodes attach to each of the two sets of storage, ostensibly in two physical locations, such as different buildings on the same campus or different metropolitan datacenters. The replicated storage can be either cluster shared volumes (CSV) or role-assigned physical disk resources (PDR). Figure 3-3 presents the typical architecture used for implementing stretch cluster replication using Storage Replica. On the left is the Redmond site, where there are two servers (SR-SRV-01 and SR-SRV- 02) and shared storage (SAN, JBOD, or iSCSI). On the right is the Bellevue site, where there are two more servers (SR-SRV-03 and SR-SRV-04) and more shared storage. You can use Storage Replica to combine the servers and shared storage at these two sites into a single stretched cluster by asymmetrically replicating storage from one site to the other. Figure 3-3: Typical architecture used for implementing stretch cluster replication This configuration makes a failover cluster tolerant not just of node failures, but entire site failures. When a single node in a site fails, another node in that site becomes the new source of replication. When all nodes in a site fail, a node in the other site becomes the source of replication. All of this occurs automatically, just like a normal nonstretched cluster. Stretch clustering requires a minimum of two nodes, and the cluster can contain up to 64 nodes. In Windows Server 2016 Technical Preview, the two cluster roles recommended for replication are Hyper-V and General Use File Server. You should avoid configuring Scale-Out File Server as a stretch cluster because Windows Server failover clusters are not inherently site aware and applications will end up connecting to nodes in both sites and then redirecting back to the owning node where I/O writes occur. This potentially can lead to poor application performance. Microsoft supports the use of virtual machine (VM) guest clusters in the Technical Preview for evaluation purposes only.

Articles in this issue

Archives of this issue

view archives of Netherlands: Software - Introductie Windows Server 2016