Overflowing workload to scalable clusters with IBM Spectrum Symphony multicluster

A scalable cluster is one that can expand (that is, scales out to be larger with more cores), so that workflow can overflow to it. The scalable cluster can offer additional resources to help or offload an application requiring more resources to complete workflow. This scalable cluster can be an on-premises private cluster, or a public cluster on the cloud. For a cloud cluster, the multicluster administrator can scale the IBM Spectrum Symphony multicluster environment using the IBM Spectrum Symphony host factory feature to leverage cloud bursting for overflow workload.

To enable scalable clusters and set the maximum cores for the cluster, a multicluster administrator configures cluster properties, by using the -s and -L parameters in the smcadmin cluster modify command, or by using the Scalable and Core Limits options in the multicluster management console.

Once enabled, you can set a cluster ranking system in your workload placement policy so that multicluster workload placement policy logic distributes workload in the order of your ranked clusters:
  1. First, the logic uses the cluster ranked first in the policy. For example, if you have an on-premises clusters ranked first, and a cloud cluster ranked second, then the logic uses the on-premises cluster first.
  2. If resources for the application are fully utilized on the first cluster, the cluster full condition for the application is therefore exceeded, and then the logic uses the second-ranked cluster (for example, the cloud cluster), which will scale-out to accommodate new overflow workload from the first cluster.

A scalable cluster allows the workload placement policy to select a scalable cluster even if its cluster full condition is met. Scalable clusters, which meet their cluster full condition, are ranked lowest of all available clusters to which the IBM Spectrum Symphony client can connect.