Traditional 2 or 3-node clusters do not provide sufficient redundancy for such independent workloads as in a Virtual Desktop Infrastructure. Additionally, data resiliency and fault tolerance capabilities become more and more important as the loss of one or a couple of nodes should not result in a failure of the whole cluster. The next critical issue for every IT infrastructure is maintaining high performance while achieving fault tolerance.
StarWind aims to resolve these issues by allowing the system to scale on Grid. Grid storage itself is an approach to storing data utilizing multiple interconnected nodes so that they can communicate with each other without switches involved.
Considering scalability, StarWind Virtual SAN provides two scaling options. Scaling-up allows adding individual hardware components, such as RAM modules, disks, flash to each node according to the current needs. However, this approach has physical limits, since when all disk, CPU and RAM slots are occupied it’s getting practically impossible to increase the capacity or performance by adding individual components.
Scale-out approach solves this problem. The main idea here is to increase both storage and compute power by adding additional nodes. This improves the reliability of the entire system since the data can be spread across multiple nodes. The storage of the scale out node is added to the existing pool or in the case of replication between two nodes, it is broken into parts with each part replicated to the partner node, thus, scaling the system on Grid. The greatest benefit of this approach is data locality, when VMs’ roles or applications have the shortest way to the storage, thus operating with considerably lower latency and as a result, have high performance.
In a Grid Architecture, every LUN is mirrored between two neighboring nodes so that each of them contents consistent data which is synchronizing permanently. This provides high performance and high resiliency. Such system can easily tolerate losing up to 50% of the nodes without any downtime at all.
The video further demonstrates the Failover Cluster of five nodes configured using StarWind Grid Architecture.