Capacity and Scaling Considerations

Understanding how failures occur will help you determine the number of on-premise search nodes you will need in your HA deployment.

There are two types of on-premise search failures that can occur:

Search Failure: In this case, new search requests will not be serviced during an outage. To design your Jive configuration to guard against this, you'll need to have more than one search broker and have them on separate nodes. This also means that you'll need to deploy an ingress replicator, which can be co-located with each of the search brokers.
Ingress Failure: In this case, new content will not be indexed during an outage. To design your Jive configuration to guard against this, you'll need to deploy multiple ingress replicators and on separate nodes. Each can be co-located with a search broker, if desired. Note that the protection against index failure is not absolute; for example, while new content continues to be indexed during an outage, a small amount of content that was created just before the outage could fail to be indexed until the ingress replicator comes back online.

Generally speaking, it makes sense to keep capacity considerations separate from your decision about your HA search configuration. Having additional search brokers adds capacity to services search requests, but a deployment would need to be very large and very active before a single search broker could not handle the requests. In that case, it would be simpler to add a CPU rather than a new search broker.

The key capacity consideration is the amount of memory available to the search brokers. Remember that the data is not shared, so each search broker needs to have enough memory to effectively handle the size of the index. Therefore, if HA is not needed, adding a second search broker for the purpose of scaling is a big investment because you would need to commit more memory to it.

For capacity planning guidelines of your Jive configuration, see Deployment Sizing and Capacity Planning.