Memory sharing among nodes in a cluster

Question

Hi,
Let's say I have two nodes A and B in the cluster. Node A has defined in config file and is using just cacheA, while node B has defined in config file and is using just cacheB.
From the fact that those two nodes are in the same cluster, does it mean that memory on nodeA is used for both cacheA and cacheB? Is memory on a node used only for caches, that are locally defined in config files and used by that node?
The reason I am asking is the following: I have many nodes, each of them has some caches defined and is using some caches, but I am expecting that some of them will be fairly small, while other big. It would be great, if from the fact, that they are in the same cluster it meant, that each node participates with it's resources to all caches in the cluster.
Best regards
Jarek

Answer 1

Marie,
it is not possible to dedicate an "own node" to specific caches in a cache service which has other caches.
To dedicate certain nodes to certain caches, you have to separate those caches to their own distributed cache service which would only be storage-enabled on the dedicated nodes. For this, you also have to explicitly configure whether that cache service is storage-enabled or storage-disabled independently of the cluster-wide tangosol.coherence.distributed.localstorage override.
Also, high-units does not tell you when you ran out of space. It tells you when you exceeded an configuration value, but that configuration value may not always correctly correspond to the actual free capacity in the JVM.
Also, high-units does not define cache capacity, which is service-wide (all storage-enabled nodes in the service contribute to it). It defines backing map capacity, which is per-node, and unfortunately the behaviour is not deterministic (unless you measured the backing map usage and defined the capacity per partition, which I believe is possible with partition-split backing maps only). Due to this the high-units does not correspond to the capacity and exceeding that capacity across the entire cache service.
Also, high-units paired with an eviction-policy is a dangerous beast on a backing map, as it will possibly lead to data loss.
Best regards,
Robert