IDUG EMEA 2024

IBM provides us number of solution to achieve multisite HA/DR for PureScale clusters. Those include technologies like HADR Replication for PureScale Clusters, Geographically Dispersed PureScale Clusters, GPFS replication for PureScale Clusters, Q Replication for PureScale Clusters.
In the presentation we will describe alternative approach, where replication across 2 sites is provided by Storage Array Layer, which allows us to achieve additional benefits like consistency groups going beyond single GPFS filesystem used by the PureScale, optimized network utilization where Storage Array replication happens on SAN Fiber, additional HA layer where filesystem can be cross mounted across sites which gives additional redundancy layer for storage availability.
 
We implement solution by deploying Synchronic Storage Replication across 2 sites, where SAN fabric provides us NSD multi path connections from each side. This way on each side we have collection of NSD disks that are local to given side, and also connection to remote NSD disks from DR side. On both sites we install standalone PureScale clusters, independent from each, therefore failure of any of those clusters does not impact other side counterpart. Having 2 clusters we designate one as primary by creating PureScale managed GPFS on top of Storage Array Layer, and on top of that create Database with its' associated paths. Next we use gpfssync capabilities to export primary GPFS filesystem , and next mount the same on DR site. This allows us to mimic primary cluster on DR site, and next attach those synchronized mount points to TSAMP cluster on secondary site. Later on, we just move application connection to DR site, which can handle further ongoing workloads.
 
Proposed solution is active-passive deployment, when at any given time only one cluster has connection to synchronized storage array. We achieve this by protections on each layer (NSD disks active on one site only, GPFS mounts online on one site only, database cataloging on one site only). This is crucial to ensure only one side has active writes to the cluster, so that we prevent any corruption of the data. In the presentation we provide procedures to setup such deployment, along with failover procedures where we move storage attachment from primary to secondary, and reverse. We also provide testing results where we verify solution, by forcing failover and more failure scenarios while processing application workload.