Separate Cache Host Process
Previously, the NCache Service and all cache instances were confined within one single process. This meant that if the process crashed, the service along with the caches and their resource information (memory, addresses, ports) were lost.
NCache has now provided more reliability by dedicating a separate process to the service and a devoted process to each cache host.
Client Communication with the Cache
The initial communication of the client with a cache host is a 2-hop communication:
It first interacts with the service, which connects the client to the cache host using the port forwarding mechanism.
It then communicates with the cache using port forwarding. All future interactions for the client are made directly with the cache.
When a cache is started in its own separate process, it is assigned a Management Port on which the service will communicate with it. This port is dynamically generated from a range (default 8300 – 8400, configurable in service.exe.config).
Note
Note that in case of the Partition-Replica topology, two ports will be assigned.
Once the port has been forwarded, the client will send all cache requests to the cache, like adding items to the cache, or fetching or removing them from the cache.
In case one of the cache host processes stops, the ports being used by that cache for communication are added back to the pool of available ports. For example, if a Partition-Replica node was using ports 8301 and 8302 and the process terminates, the ports will become available for use by any other cache. Now when a process starts for a new cache, port 8301 can be reused for this cache.
Formerly, once a port was assigned to the cache, it was considered as utilized - regardless of the cache state. This meant the port range was narrowed down involuntarily.
See Also
Graceful Node Down
Self Healing Dynamic Clustering
IP Binding with Multiple NICs
Split Brain
