Heap size does not directly affect throughput but it is affected by the message
size and the number of messages being processed concurrently. Since an estimation of the size
of messages that is being processed concurrently is very difficult, you should try to make a
worst case estimate by taking the resulting XML file size, multiply it with 20, and multiply
it with the number of processing threads.
Heap estimate applies for the Central and Process node types and to MapGen and
UI. MapGen and UI will suffice with less heap. MapGen depends on the number of preload threads
and UI depends on the number of users and the type of operations performed. The MapGen and UI
node types require a 256 to 1024MB heap.
Messages that have streaming maps do not affect the heap size for the Central
or Process nodes. Messages that result in large XML files should have streaming mappings,
and streaming requires that the map is constructed to support this. The worst case scenario
does not apply to ordered messages with the same primary key, since only one primary key is
processed at a time.